--- /dev/null
+
+# dyncnn is a deep-learning algorithm for the prediction of
+# interacting object dynamics
+#
+# Copyright (c) 2016 Idiap Research Institute, http://www.idiap.ch/
+# Written by Francois Fleuret <francois.fleuret@idiap.ch>
+#
+# This file is part of dyncnn.
+#
+# dyncnn is free software: you can redistribute it and/or modify it
+# under the terms of the GNU General Public License version 3 as
+# published by the Free Software Foundation.
+#
+# dyncnn is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+# General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with dyncnn. If not, see <http://www.gnu.org/licenses/>.
+
+ifeq ($(DEBUG),yes)
+ CXXFLAGS = -fPIC -Wall -g -DDEBUG
+else
+ CXXFLAGS = -fPIC -Wall -g -O3
+endif
+
+CXXFLAGS += -I/usr/include/cairo -DCAIRO_SUPPORT
+
+LDFLAGS += -lcairo
+
+all: flatland.so TAGS
+
+TAGS: *.cc *.h
+ etags *.cc *.h
+
+flatland.so: flatland.o misc.o \
+ universe.o \
+ polygon.o \
+ canvas.o canvas_cairo.o
+ $(CXX) $(CXXFLAGS) $(LDFLAGS) -shared -fPIC -o $@ $^
+
+test: test.c flatland.so
+ $(CC) flatland.so -o test test.c
+
+Makefile.depend: *.h *.cc Makefile
+ $(CC) $(CXXFLAGS) -M *.cc > Makefile.depend
+
+clean:
+ \rm -f flatland *.o *.so Makefile.depend
+
+-include Makefile.depend
--- /dev/null
+#!/usr/bin/env python-for-pytorch
+
+import os
+from torch.utils.ffi import create_extension
+
+this_file = os.path.dirname(__file__)
+
+print('__file__', __file__)
+
+ffi = create_extension(
+ '_ext.mylib',
+ headers = [ 'mylib.h' ],
+ sources = [ 'mylib.c' ],
+ extra_objects = [ '/home/fleuret/sources/python/flatland/flatland.so' ],
+ libraries = [ ],
+ library_dirs = [ ],
+ define_macros = [ ],
+ relative_to=this_file,
+ with_cuda = False
+)
+
+if __name__ == '__main__':
+ ffi.build()
--- /dev/null
+
+/*
+ * dyncnn is a deep-learning algorithm for the prediction of
+ * interacting object dynamics
+ *
+ * Copyright (c) 2016 Idiap Research Institute, http://www.idiap.ch/
+ * Written by Francois Fleuret <francois.fleuret@idiap.ch>
+ *
+ * This file is part of dyncnn.
+ *
+ * dyncnn is free software: you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 3 as
+ * published by the Free Software Foundation.
+ *
+ * dyncnn is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with dyncnn. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#include "canvas.h"
+
+Canvas::~Canvas() {}
--- /dev/null
+
+/*
+ * dyncnn is a deep-learning algorithm for the prediction of
+ * interacting object dynamics
+ *
+ * Copyright (c) 2016 Idiap Research Institute, http://www.idiap.ch/
+ * Written by Francois Fleuret <francois.fleuret@idiap.ch>
+ *
+ * This file is part of dyncnn.
+ *
+ * dyncnn is free software: you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 3 as
+ * published by the Free Software Foundation.
+ *
+ * dyncnn is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with dyncnn. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#ifndef CANVAS_H
+#define CANVAS_H
+
+#include "misc.h"
+
+class Canvas {
+public:
+ virtual ~Canvas();
+ virtual void clear() = 0;
+ virtual void set_line_width(scalar_t w) = 0;
+ virtual void set_drawing_color(scalar_t r, scalar_t g, scalar_t b) = 0;
+ virtual void draw_polygon(int filled, int nb, scalar_t *x, scalar_t *y) = 0;
+};
+
+#endif
--- /dev/null
+
+/*
+ * dyncnn is a deep-learning algorithm for the prediction of
+ * interacting object dynamics
+ *
+ * Copyright (c) 2016 Idiap Research Institute, http://www.idiap.ch/
+ * Written by Francois Fleuret <francois.fleuret@idiap.ch>
+ *
+ * This file is part of dyncnn.
+ *
+ * dyncnn is free software: you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 3 as
+ * published by the Free Software Foundation.
+ *
+ * dyncnn is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with dyncnn. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#include "canvas_cairo.h"
+
+#include <cmath>
+
+#define MAX(x, y) ((x >= y) ? (x) : (y))
+
+CanvasCairo::CanvasCairo(scalar_t scale, int nb_rows, int nb_cols, CanvasCairo **ca) {
+ _actual_width = 0;
+ _actual_height = 0;
+
+ for(int i = 0; i < nb_rows; i++) {
+ int row_height = 0, row_width = 0;
+ for(int j = 0; j < nb_cols; j++) {
+ CanvasCairo *this_ca = ca[i * nb_cols + j];
+ row_height = MAX(row_height, this_ca->_actual_height);
+ row_width += this_ca->_actual_width;
+ }
+ _actual_width = MAX(_actual_width, row_width);
+ _actual_height += row_height;
+ }
+
+ _data = new unsigned char [_actual_width * _actual_height * _depth];
+
+ int x0, y0 = 0;
+ for(int i = 0; i < nb_rows; i++) {
+ x0 = 0;
+ int row_height = 0;
+ for(int j = 0; j < nb_cols; j++) {
+ CanvasCairo *this_ca = ca[i * nb_cols + j];
+ for(int y = 0; y < this_ca->_actual_height; y++) {
+ for(int x = 0; x < this_ca->_actual_width; x++) {
+ for(int d = 0; d < _depth; d++) {
+ _data[(x0 + x + _actual_width * (y0 + y))* _depth + d] =
+ this_ca->_data[(x + this_ca->_actual_width * y)* _depth + d];
+
+ }
+ }
+ }
+ row_height = MAX(row_height, this_ca->_actual_height);
+ x0 += this_ca->_actual_width;
+ }
+ y0 += row_height;
+ }
+
+ _image = cairo_image_surface_create_for_data(_data,
+ CAIRO_FORMAT_RGB24,
+ _actual_width,
+ _actual_height,
+ _actual_width * _depth);
+
+ _context_resource = cairo_create(_image);
+
+ cairo_scale(_context_resource, scale, scale);
+
+ clear();
+ // cairo_set_source_rgb(_context_resource, 1.0, 1.0, 1.0);
+ // cairo_set_line_width (_context_resource, 1.0);
+}
+
+CanvasCairo::CanvasCairo(scalar_t scale, int width, int height) {
+ _actual_width = int(width * scale);
+ _actual_height = int(height * scale);
+ _scale = scale;
+
+ _data = new unsigned char [_actual_width * _actual_height * _depth];
+
+ _image = cairo_image_surface_create_for_data(_data,
+ CAIRO_FORMAT_RGB24,
+ _actual_width,
+ _actual_height,
+ _actual_width * _depth);
+
+ _context_resource = cairo_create(_image);
+
+ cairo_scale(_context_resource, scale, scale);
+
+ cairo_set_line_width (_context_resource, 1.0);
+
+ clear();
+ // cairo_set_source_rgb(_context_resource, 1.0, 1.0, 1.0);
+ // cairo_rectangle(_context_resource, 0, 0, width, height);
+ // cairo_fill(_context_resource);
+}
+
+CanvasCairo::~CanvasCairo() {
+ cairo_destroy(_context_resource);
+ cairo_surface_destroy(_image);
+ delete[] _data;
+}
+
+void CanvasCairo::clear() {
+ cairo_set_source_rgb(_context_resource, 1.0, 1.0, 1.0);
+ cairo_rectangle(_context_resource, 0, 0, _actual_width / _scale, _actual_height / _scale);
+ cairo_fill(_context_resource);
+}
+
+void CanvasCairo::set_line_width(scalar_t w) {
+ cairo_set_line_width (_context_resource, w);
+}
+
+void CanvasCairo::set_drawing_color(scalar_t r, scalar_t g, scalar_t b) {
+ cairo_set_source_rgb(_context_resource, r, g, b);
+}
+
+void CanvasCairo::draw_polygon(int filled, int nb, scalar_t *x, scalar_t *y) {
+ cairo_move_to(_context_resource, x[0], y[0]);
+ for(int n = 0; n < nb; n++) {
+ cairo_line_to(_context_resource, x[n], y[n]);
+ }
+ cairo_close_path(_context_resource);
+
+ if(filled) {
+ cairo_stroke_preserve(_context_resource);
+ cairo_fill(_context_resource);
+ } else {
+ cairo_stroke(_context_resource);
+ }
+}
+
+static cairo_status_t write_cairo_to_file(void *closure,
+ const unsigned char *data,
+ unsigned int length) {
+ fwrite(data, 1, length, (FILE *) closure);
+ return CAIRO_STATUS_SUCCESS;
+}
+
+void CanvasCairo::write_png(FILE *file) {
+ cairo_surface_write_to_png_stream(_image, write_cairo_to_file, file);
+}
--- /dev/null
+
+/*
+ * dyncnn is a deep-learning algorithm for the prediction of
+ * interacting object dynamics
+ *
+ * Copyright (c) 2016 Idiap Research Institute, http://www.idiap.ch/
+ * Written by Francois Fleuret <francois.fleuret@idiap.ch>
+ *
+ * This file is part of dyncnn.
+ *
+ * dyncnn is free software: you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 3 as
+ * published by the Free Software Foundation.
+ *
+ * dyncnn is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with dyncnn. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#ifndef CANVAS_CAIRO_H
+#define CANVAS_CAIRO_H
+
+#include "canvas.h"
+
+#include <stdio.h>
+#include <cairo.h>
+
+class CanvasCairo : public Canvas {
+ scalar_t _scale;
+ cairo_surface_t *_image;
+ cairo_t *_context_resource;
+
+public:
+ const static int _depth = 4;
+ int _actual_width, _actual_height;
+ unsigned char *_data;
+
+ CanvasCairo(scalar_t scale, int nb_rows, int nb_cols, CanvasCairo **ca);
+ CanvasCairo(scalar_t scale, int width, int height);
+ CanvasCairo(int nb_rows, int nb_cols, CanvasCairo **x);
+
+ virtual ~CanvasCairo();
+
+ virtual void clear();
+ virtual void set_line_width(scalar_t w);
+ virtual void set_drawing_color(scalar_t r, scalar_t g, scalar_t b);
+ virtual void draw_polygon(int filled, int nb, scalar_t *x, scalar_t *y);
+ virtual void write_png(FILE *file);
+};
+
+#endif
--- /dev/null
+
+/*
+ * dyncnn is a deep-learning algorithm for the prediction of
+ * interacting object dynamics
+ *
+ * Copyright (c) 2016 Idiap Research Institute, http://www.idiap.ch/
+ * Written by Francois Fleuret <francois.fleuret@idiap.ch>
+ *
+ * This file is part of dyncnn.
+ *
+ * dyncnn is free software: you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 3 as
+ * published by the Free Software Foundation.
+ *
+ * dyncnn is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with dyncnn. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#include <iostream>
+#include <fstream>
+#include <cmath>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <errno.h>
+#include <string.h>
+#include <sys/stat.h>
+#include <sys/time.h>
+
+using namespace std;
+
+#include "misc.h"
+#include "universe.h"
+#include "canvas_cairo.h"
+
+FILE *safe_fopen(const char *name, const char *mode) {
+ FILE *file = fopen(name, mode);
+ if(!file) {
+ cerr << "Cannot open " << name << endl;
+ exit(1);
+ }
+ return file;
+}
+
+void print_help(const char *command) {
+ cerr << command << " <nb sequences to generate> [--dir <dir>] [--seed <seed>]]" << endl;
+ exit(1);
+}
+
+//////////////////////////////////////////////////////////////////////
+
+void draw_universe_on_canvas(CanvasCairo *canvas, scalar_t scaling,
+ Universe *universe) {
+ canvas->set_line_width(1.0 / scaling);
+ universe->draw(canvas);
+}
+
+void draw_grabbing_point_on_canvas(CanvasCairo *canvas, scalar_t scaling,
+ scalar_t xg, scalar_t yg,
+ scalar_t r, scalar_t g, scalar_t b) {
+ scalar_t radius = 1/scaling;
+ int n = 36;
+ scalar_t xp[n], yp[n];
+ for(int k = 0; k < n; k++) {
+ scalar_t alpha = 2 * M_PI * scalar_t(k) / scalar_t(n);
+ xp[k] = xg + radius * cos(alpha);
+ yp[k] = yg + radius * sin(alpha);
+ }
+ canvas->set_drawing_color(r, g, b);
+ canvas->set_line_width(2.0);
+ canvas->draw_polygon(1, n, xp, yp);
+}
+
+//////////////////////////////////////////////////////////////////////
+
+extern "C" void fl_generate_sequences(int nb_sequences,
+ int nb_images_per_sequence,
+ int width, int height,
+ unsigned char *output) {
+
+ const scalar_t world_width = width * 8;
+ const scalar_t world_height = height * 8;
+ const scalar_t scaling = 0.125;
+
+ const scalar_t dt = 0.1;
+ const int nb_iterations_per_steps = 5;
+
+ //////////////////////////////////////////////////////////////////////
+
+ // We will generate images { 0, every_nth, 2 * every_nth, ..., k * every_nth < nb_simulated_frames }
+
+ // The framerate every_nth may be set to smaller value to generate
+ // nice materials for presentations or papers.
+
+ int every_nth = 16;
+ int nb_simulated_frames = 1 + (nb_images_per_sequence - 1) * every_nth;
+ int random_grasp = 1;
+ int random_shape_size = 0;
+ int nb_shapes = 10;
+ // char data_dir[1024] = "/tmp/";
+ // int show_grabbing_point = 0;
+ int skip = -1;
+
+ for(int n = 0; n < nb_sequences; n++) {
+
+ Universe *universe;
+ Polygon *grabbed_polygon;
+
+ universe = new Universe(nb_shapes, world_width, world_height);
+
+ const int nb_saved_frames = (nb_simulated_frames + every_nth - 1) / every_nth;
+ if(nb_saved_frames != nb_images_per_sequence) {
+ cerr << "It makes no sense." << endl;
+ abort();
+ }
+
+ CanvasCairo *canvases[nb_saved_frames * 2];
+
+ for(int s = 0; s < 2 * nb_saved_frames; s++) {
+ canvases[s] = new CanvasCairo(scaling, universe->width(), universe->height());
+ }
+
+ scalar_t grab_start_x, grab_start_y;
+
+ if(random_grasp) {
+ grab_start_x = world_width * (0.1 + 0.8 * drand48());
+ grab_start_y = world_height * (0.1 + 0.8 * drand48());
+ } else {
+ grab_start_x = world_width * 0.5;
+ grab_start_y = world_height * 0.75;
+ }
+
+ do {
+ universe->clear();
+
+ const int nb_attempts_max = 100;
+ int nb_attempts = 0;
+
+ for(int u = 0; u < nb_shapes; u++) {
+ Polygon *pol = 0;
+
+ nb_attempts = 0;
+
+ scalar_t shape_size;
+
+ if(random_shape_size) {
+ shape_size = 40 + 80 * drand48();
+ } else {
+ shape_size = 80;
+ }
+
+ do {
+ scalar_t x[] = { - shape_size * 0.4, + shape_size * 0.4,
+ + shape_size * 0.4, - shape_size * 0.4 };
+
+ scalar_t y[] = { - shape_size * 0.6, - shape_size * 0.6,
+ + shape_size * 0.6, + shape_size * 0.6 };
+
+ scalar_t delta = shape_size / sqrt(2.0);
+
+ scalar_t object_center_x = delta + (world_width - 2 * delta) * drand48();
+ scalar_t object_center_y = delta + (world_height - 2 * delta) * drand48();
+
+ delete pol;
+ pol = new Polygon(0.5, 1.0, 1.0, 1.0, x, y, sizeof(x)/sizeof(scalar_t));
+ pol->set_position(object_center_x, object_center_y, M_PI * 2 * drand48());
+ pol->set_speed(0, 0, 0);
+
+ universe->initialize_polygon(pol);
+
+ nb_attempts++;
+ } while(nb_attempts < nb_attempts_max && universe->collide(pol));
+
+ if(nb_attempts == nb_attempts_max) {
+ delete pol;
+ u = -1;
+ universe->clear();
+ nb_attempts = 0;
+ } else {
+ universe->add_polygon(pol);
+ }
+ }
+
+ grabbed_polygon = universe->pick_polygon(grab_start_x, grab_start_y);
+ } while(!grabbed_polygon);
+
+ if(skip < 0 || n >= skip) {
+
+ scalar_t grab_relative_x = grabbed_polygon->relative_x(grab_start_x, grab_start_y);
+ scalar_t grab_relative_y = grabbed_polygon->relative_y(grab_start_x, grab_start_y);
+
+ for(int s = 0; s < nb_simulated_frames; s++) {
+ if(s % every_nth == 0) {
+ int t = s / every_nth;
+ // scalar_t xf = grabbed_polygon->absolute_x(grab_relative_x, grab_relative_y);
+ // scalar_t yf = grabbed_polygon->absolute_y(grab_relative_x, grab_relative_y);
+
+ // canvases[2 * t + 0]->clear();
+ // draw_grabbing_point_on_canvas(canvases[2 * t + 0], scaling,
+ // xf, yf, 0.0, 0.0, 0.0);
+ // canvases[2 * t + 1]->clear();
+ // draw_universe_on_canvas(canvases[2 * t + 1], scaling, universe);
+
+ canvases[t]->clear();
+ draw_universe_on_canvas(canvases[t], scaling, universe);
+
+ // if(show_grabbing_point) {
+ // draw_grabbing_point_on_canvas(canvases[2 * t + 1], scaling,
+ // xf, yf, 1.0, 0.0, 0.0);
+ // }
+ }
+
+ if(s < nb_simulated_frames - 1) {
+ // Run the simulation
+ for(int i = 0; i < nb_iterations_per_steps; i++) {
+ scalar_t xf = grabbed_polygon->absolute_x(grab_relative_x, grab_relative_y);
+ scalar_t yf = grabbed_polygon->absolute_y(grab_relative_x, grab_relative_y);
+ grabbed_polygon->apply_force(dt, xf, yf, 0.0, -1.0);
+ universe->update(dt);
+ }
+ }
+ }
+
+ for(int t = 0; t < nb_images_per_sequence; t++) {
+ memcpy(output + (n * nb_images_per_sequence + t) * 3 * width * height,
+ canvases[t]->_data,
+ 3 * width * height);
+ }
+
+ // char buffer[1024];
+ // sprintf(buffer, "%s/%03d/dyn_%06d.png", data_dir, n / 1000, n);
+ // FILE *file = safe_fopen(buffer, "w");
+ // main_canvas.write_png(file);
+ // fclose(file);
+ }
+
+ for(int t = 0; t < 2 * nb_saved_frames; t++) {
+ delete canvases[t];
+ }
+
+ delete universe;
+ }
+}
--- /dev/null
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void fl_generate_sequences(int nb_sequences,
+ int nb_images_per_sequence,
+ int width, int height,
+ unsigned char *output);
+
+#ifdef __cplusplus
+}
+#endif
+
--- /dev/null
+
+/*
+ * dyncnn is a deep-learning algorithm for the prediction of
+ * interacting object dynamics
+ *
+ * Copyright (c) 2016 Idiap Research Institute, http://www.idiap.ch/
+ * Written by Francois Fleuret <francois.fleuret@idiap.ch>
+ *
+ * This file is part of dyncnn.
+ *
+ * dyncnn is free software: you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 3 as
+ * published by the Free Software Foundation.
+ *
+ * dyncnn is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with dyncnn. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#include "misc.h"
+
+int compare_couple(const void *a, const void *b) {
+ if(((Couple *) a)->value < ((Couple *) b)->value) return -1;
+ else if(((Couple *) a)->value > ((Couple *) b)->value) return 1;
+ else return 0;
+}
--- /dev/null
+
+/*
+ * dyncnn is a deep-learning algorithm for the prediction of
+ * interacting object dynamics
+ *
+ * Copyright (c) 2016 Idiap Research Institute, http://www.idiap.ch/
+ * Written by Francois Fleuret <francois.fleuret@idiap.ch>
+ *
+ * This file is part of dyncnn.
+ *
+ * dyncnn is free software: you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 3 as
+ * published by the Free Software Foundation.
+ *
+ * dyncnn is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with dyncnn. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#ifndef MISC_H
+#define MISC_H
+
+#include <float.h>
+#include <stdlib.h>
+
+#ifdef DEBUG
+#define ASSERT(x, s) if(!(x)) { std::cerr << "ASSERT FAILED IN " << __FILE__ << ":" << __LINE__ << " [" << (s) << "]\n"; abort(); }
+#else
+#define ASSERT(x, s)
+#endif
+
+// typedef float scalar_t;
+typedef double scalar_t;
+
+inline scalar_t sq(scalar_t x) { return x*x; }
+
+inline scalar_t prod_vect(scalar_t x1, scalar_t y1, scalar_t x2, scalar_t y2) {
+ return x1 * y2 - x2 * y1;
+}
+
+struct Couple {
+ int index;
+ scalar_t value;
+};
+
+int compare_couple(const void *a, const void *b);
+
+#endif
--- /dev/null
+
+#include <TH/TH.h>
+
+/*
+
+ Example of FFI extension I started from:
+
+ https://github.com/pytorch/extension-ffi.git
+
+ There is this tutorial
+
+ https://github.com/pytorch/tutorials/blob/master/Creating%20Extensions%20using%20FFI.md
+
+ And TH's Tensor definition are here in my install:
+
+ anaconda3/lib/python3.5/site-packages/torch/lib/include/TH/generic/THTensor.h
+
+ */
+
+#include "flatland.h"
+
+int generate_sequence(THByteTensor *output) {
+ long nb_sequences = 1;
+ long nb_images_per_sequence = 5;
+ long depth = 3;
+ long width = 64;
+ long height = 64;
+ long s;
+ unsigned char *a, *b;
+ int c, k, i, j, st0, st1, st2, st3, st4;
+
+ THByteTensor_resize5d(output, nb_sequences, nb_images_per_sequence, depth, height, width);
+
+ st0 = THByteTensor_stride(output, 0);
+ st1 = THByteTensor_stride(output, 1);
+ st2 = THByteTensor_stride(output, 2);
+ st3 = THByteTensor_stride(output, 3);
+ st4 = THByteTensor_stride(output, 4);
+
+ a =
+ THByteTensor_storage(output)->data + THByteTensor_storageOffset(output);
+
+ for(s = 0; s < nb_sequences; s++) {
+ unsigned char result[nb_images_per_sequence * depth * width * height];
+ unsigned char *r = result;
+ fl_generate_sequences(1, nb_images_per_sequence, width, height, result);
+ for(k = 0; k < nb_images_per_sequence; k++) {
+ for(c = 0; c < depth; c++) {
+ for(i = 0; i < height; i++) {
+ b = a
+ + s * st0 + k * st1 + c * st2 + i * st3;
+ for(j = 0; j < width; j++) {
+ *b = (unsigned char) (*r);
+ r++;
+ b += st4;
+ }
+ }
+ }
+ }
+ }
+
+ return 1;
+}
--- /dev/null
+
+int generate_sequence(THByteTensor *output);
--- /dev/null
+
+/*
+ * dyncnn is a deep-learning algorithm for the prediction of
+ * interacting object dynamics
+ *
+ * Copyright (c) 2016 Idiap Research Institute, http://www.idiap.ch/
+ * Written by Francois Fleuret <francois.fleuret@idiap.ch>
+ *
+ * This file is part of dyncnn.
+ *
+ * dyncnn is free software: you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 3 as
+ * published by the Free Software Foundation.
+ *
+ * dyncnn is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with dyncnn. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#include <iostream>
+
+using namespace std;
+
+#include <cmath>
+#include "polygon.h"
+
+static const scalar_t dl = 20.0;
+static const scalar_t repulsion_constant = 0.2;
+static const scalar_t dissipation = 0.5;
+
+Polygon::Polygon(scalar_t mass,
+ scalar_t red, scalar_t green, scalar_t blue,
+ scalar_t *x, scalar_t *y,
+ int nv) : _mass(mass),
+ _moment_of_inertia(0), _radius(0),
+ _relative_x(new scalar_t[nv]), _relative_y(new scalar_t[nv]),
+ _total_nb_dots(0),
+ _nb_dots(new int[nv]),
+ _effecting_edge(0),
+ _length(new scalar_t[nv]),
+ _triangles(new Triangle[nv-2]),
+ _initialized(false), _nailed(false),
+ _nb_vertices(nv),
+ _x(new scalar_t[nv]), _y(new scalar_t[nv]),
+ _red(red), _green(green), _blue(blue) {
+ _last_center_x = 0;
+ _last_center_y = 0;
+ _last_theta = 0;
+
+ if(x) for(int i = 0; i < nv; i++) _relative_x[i] = x[i];
+ if(y) for(int i = 0; i < nv; i++) _relative_y[i] = y[i];
+}
+
+Polygon::~Polygon() {
+ delete[] _relative_x;
+ delete[] _relative_y;
+ delete[] _x;
+ delete[] _y;
+ delete[] _nb_dots;
+ delete[] _length;
+ delete[] _triangles;
+ delete[] _effecting_edge;
+}
+
+Polygon *Polygon::clone() {
+ return new Polygon(_mass, _red, _green, _blue, _relative_x, _relative_y, _nb_vertices);
+}
+
+#ifdef XFIG_SUPPORT
+void Polygon::color_xfig(XFigTracer *tracer) {
+ tracer->add_color(int(255 * _red), int(255 * _green), int(255 * _blue));
+}
+
+void Polygon::print_xfig(XFigTracer *tracer) {
+ tracer->draw_polygon(int(255 * _red), int(255 * _green), int(255 * _blue),
+ _nb_vertices, _x, _y);
+}
+#endif
+
+#ifdef X11_SUPPORT
+void Polygon::draw(SimpleWindow *window) {
+ window->color(_red, _green, _blue);
+ int x[_nb_vertices], y[_nb_vertices];
+ for(int n = 0; n < _nb_vertices; n++) {
+ x[n] = int(_x[n]);
+ y[n] = int(_y[n]);
+ }
+ window->fill_polygon(_nb_vertices, x, y);
+}
+
+void Polygon::draw_contours(SimpleWindow *window) {
+ int x[_nb_vertices], y[_nb_vertices];
+ for(int n = 0; n < _nb_vertices; n++) {
+ x[n] = int(_x[n]);
+ y[n] = int(_y[n]);
+ }
+ window->color(0.0, 0.0, 0.0);
+ // window->color(1.0, 1.0, 1.0);
+ for(int n = 0; n < _nb_vertices; n++) {
+ window->draw_line(x[n], y[n], x[(n+1)%_nb_vertices], y[(n+1)%_nb_vertices]);
+ }
+}
+#endif
+
+void Polygon::draw(Canvas *canvas) {
+ canvas->set_drawing_color(_red, _green, _blue);
+ canvas->draw_polygon(1, _nb_vertices, _x, _y);
+}
+
+void Polygon::draw_contours(Canvas *canvas) {
+ canvas->set_drawing_color(0.0, 0.0, 0.0);
+ canvas->draw_polygon(0, _nb_vertices, _x, _y);
+}
+
+void Polygon::set_vertex(int k, scalar_t x, scalar_t y) {
+ _relative_x[k] = x;
+ _relative_y[k] = y;
+}
+
+void Polygon::set_position(scalar_t center_x, scalar_t center_y, scalar_t theta) {
+ _center_x = center_x;
+ _center_y = center_y;
+ _theta = theta;
+}
+
+void Polygon::set_speed(scalar_t dcenter_x, scalar_t dcenter_y, scalar_t dtheta) {
+ _dcenter_x = dcenter_x;
+ _dcenter_y = dcenter_y;
+ _dtheta = dtheta;
+}
+
+bool Polygon::contain(scalar_t x, scalar_t y) {
+ for(int t = 0; t < _nb_vertices-2; t++) {
+ scalar_t xa = _x[_triangles[t].a], ya = _y[_triangles[t].a];
+ scalar_t xb = _x[_triangles[t].b], yb = _y[_triangles[t].b];
+ scalar_t xc = _x[_triangles[t].c], yc = _y[_triangles[t].c];
+ if(prod_vect(x - xa, y - ya, xb - xa, yb - ya) <= 0 &&
+ prod_vect(x - xb, y - yb, xc - xb, yc - yb) <= 0 &&
+ prod_vect(x - xc, y - yc, xa - xc, ya - yc) <= 0) return true;
+ }
+ return false;
+}
+
+void Polygon::triangularize(int &nt, int nb, int *index) {
+ if(nb == 3) {
+
+ if(nt >= _nb_vertices-2) {
+ cerr << "Error type #1 in triangularization." << endl;
+ exit(1);
+ }
+
+ _triangles[nt].a = index[0];
+ _triangles[nt].b = index[1];
+ _triangles[nt].c = index[2];
+ nt++;
+
+ } else {
+ int best_m = -1, best_n = -1;
+ scalar_t best_split = -1, det, s = -1, t = -1;
+
+ for(int n = 0; n < nb; n++) for(int m = 0; m < n; m++) if(n > m+1 && m+nb > n+1) {
+ bool no_intersection = true;
+ for(int k = 0; no_intersection & (k < nb); k++)
+ if(k != n && k != m && (k+1)%nb != n && (k+1)%nb != m) {
+ intersection(_relative_x[index[n]], _relative_y[index[n]],
+ _relative_x[index[m]], _relative_y[index[m]],
+ _relative_x[index[k]], _relative_y[index[k]],
+ _relative_x[index[(k+1)%nb]], _relative_y[index[(k+1)%nb]], det, s, t);
+ no_intersection = det == 0 || s < 0 || s > 1 || t < 0 || t > 1;
+ }
+
+ if(no_intersection) {
+ scalar_t a1 = 0, a2 = 0;
+ for(int k = 0; k < nb; k++) if(k >= m && k < n)
+ a1 += prod_vect(_relative_x[index[k]] - _relative_x[index[m]],
+ _relative_y[index[k]] - _relative_y[index[m]],
+ _relative_x[index[k+1]] - _relative_x[index[m]],
+ _relative_y[index[k+1]] - _relative_y[index[m]]);
+ else
+ a2 += prod_vect(_relative_x[index[k]] - _relative_x[index[m]],
+ _relative_y[index[k]] - _relative_y[index[m]],
+ _relative_x[index[(k+1)%nb]] - _relative_x[index[m]],
+ _relative_y[index[(k+1)%nb]] - _relative_y[index[m]]);
+
+ if((a1 * a2 > 0 && best_split < 0) || (abs(a1 - a2) < best_split)) {
+ best_n = n; best_m = m;
+ best_split = abs(a1 - a2);
+ }
+ }
+ }
+
+ if(best_n >= 0 && best_m >= 0) {
+ int index_neg[nb], index_pos[nb];
+ int neg = 0, pos = 0;
+ for(int k = 0; k < nb; k++) {
+ if(k >= best_m && k <= best_n) index_pos[pos++] = index[k];
+ if(k <= best_m || k >= best_n) index_neg[neg++] = index[k];
+ }
+ if(pos < 3 || neg < 3) {
+ cerr << "Error type #2 in triangularization." << endl;
+ exit(1);
+ }
+ triangularize(nt, pos, index_pos);
+ triangularize(nt, neg, index_neg);
+ } else {
+ cerr << "Error type #3 in triangularization." << endl;
+ exit(1);
+ }
+ }
+}
+
+void Polygon::initialize(int nb_polygons) {
+ scalar_t a;
+
+ _nb_polygons = nb_polygons;
+
+ a = _relative_x[_nb_vertices - 1] * _relative_y[0]
+ - _relative_x[0] * _relative_y[_nb_vertices - 1];
+
+ for(int n = 0; n < _nb_vertices - 1; n++)
+ a += _relative_x[n] * _relative_y[n+1] - _relative_x[n+1] * _relative_y[n];
+ a *= 0.5;
+
+ // Reorder the vertices
+
+ if(a < 0) {
+ a = -a;
+ scalar_t x, y;
+ for(int n = 0; n < _nb_vertices / 2; n++) {
+ x = _relative_x[n];
+ y = _relative_y[n];
+ _relative_x[n] = _relative_x[_nb_vertices - 1 - n];
+ _relative_y[n] = _relative_y[_nb_vertices - 1 - n];
+ _relative_x[_nb_vertices - 1 - n] = x;
+ _relative_y[_nb_vertices - 1 - n] = y;
+ }
+ }
+
+ // Compute the center of mass and moment of inertia
+
+ scalar_t sx, sy, w;
+ w = 0;
+ sx = 0;
+ sy = 0;
+ for(int n = 0; n < _nb_vertices; n++) {
+ int np = (n+1)%_nb_vertices;
+ w =_relative_x[n] * _relative_y[np] - _relative_x[np] * _relative_y[n];
+ sx += (_relative_x[n] + _relative_x[np]) * w;
+ sy += (_relative_y[n] + _relative_y[np]) * w;
+ }
+ sx /= 6 * a;
+ sy /= 6 * a;
+
+ _radius = 0;
+ for(int n = 0; n < _nb_vertices; n++) {
+ _relative_x[n] -= sx;
+ _relative_y[n] -= sy;
+ scalar_t r = sqrt(sq(_relative_x[n]) + sq(_relative_y[n]));
+ if(r > _radius) _radius = r;
+ }
+
+ scalar_t num = 0, den = 0;
+ for(int n = 0; n < _nb_vertices; n++) {
+ int np = (n+1)%_nb_vertices;
+ den += abs(prod_vect(_relative_x[np], _relative_y[np], _relative_x[n], _relative_y[n]));
+ num += abs(prod_vect(_relative_x[np], _relative_y[np], _relative_x[n], _relative_y[n])) *
+ (_relative_x[np] * _relative_x[np] + _relative_y[np] * _relative_y[np] +
+ _relative_x[np] * _relative_x[n] + _relative_y[np] * _relative_y[n] +
+ _relative_x[n] * _relative_x[n] + _relative_y[n] * _relative_y[n]);
+ }
+
+ _moment_of_inertia = num / (6 * den);
+
+ scalar_t vx = cos(_theta), vy = sin(_theta);
+
+ for(int n = 0; n < _nb_vertices; n++) {
+ _x[n] = _center_x + _relative_x[n] * vx + _relative_y[n] * vy;
+ _y[n] = _center_y - _relative_x[n] * vy + _relative_y[n] * vx;
+ }
+
+ scalar_t length;
+ _total_nb_dots = 0;
+ for(int n = 0; n < _nb_vertices; n++) {
+ length = sqrt(sq(_relative_x[n] - _relative_x[(n+1)%_nb_vertices]) +
+ sq(_relative_y[n] - _relative_y[(n+1)%_nb_vertices]));
+ _length[n] = length;
+ _nb_dots[n] = int(length / dl + 1);
+ _total_nb_dots += _nb_dots[n];
+ }
+
+ delete[] _effecting_edge;
+ _effecting_edge = new int[_nb_polygons * _total_nb_dots];
+ for(int p = 0; p < _nb_polygons * _total_nb_dots; p++) _effecting_edge[p] = -1;
+
+ int nt = 0;
+ int index[_nb_vertices];
+ for(int n = 0; n < _nb_vertices; n++) index[n] = n;
+ triangularize(nt, _nb_vertices, index);
+
+ _initialized = true;
+}
+
+bool Polygon::update(scalar_t dt) {
+ if(!_nailed) {
+ _center_x += _dcenter_x * dt;
+ _center_y += _dcenter_y * dt;
+ _theta += _dtheta * dt;
+ }
+
+ scalar_t d = exp(log(dissipation) * dt);
+ _dcenter_x *= d;
+ _dcenter_y *= d;
+ _dtheta *= d;
+
+ scalar_t vx = cos(_theta), vy = sin(_theta);
+
+ for(int n = 0; n < _nb_vertices; n++) {
+ _x[n] = _center_x + _relative_x[n] * vx + _relative_y[n] * vy;
+ _y[n] = _center_y - _relative_x[n] * vy + _relative_y[n] * vx;
+ }
+
+ if(abs(_center_x - _last_center_x) +
+ abs(_center_y - _last_center_y) +
+ abs(_theta - _last_theta) * _radius > 0.1) {
+ _last_center_x = _center_x;
+ _last_center_y = _center_y;
+ _last_theta = _theta;
+ return true;
+ } else return false;
+}
+
+scalar_t Polygon::relative_x(scalar_t ax, scalar_t ay) {
+ return (ax - _center_x) * cos(_theta) - (ay - _center_y) * sin(_theta);
+}
+
+scalar_t Polygon::relative_y(scalar_t ax, scalar_t ay) {
+ return (ax - _center_x) * sin(_theta) + (ay - _center_y) * cos(_theta);
+}
+
+scalar_t Polygon::absolute_x(scalar_t rx, scalar_t ry) {
+ return _center_x + rx * cos(_theta) + ry * sin(_theta);
+}
+
+scalar_t Polygon::absolute_y(scalar_t rx, scalar_t ry) {
+ return _center_y - rx * sin(_theta) + ry * cos(_theta);
+}
+
+void Polygon::apply_force(scalar_t dt, scalar_t x, scalar_t y, scalar_t fx, scalar_t fy) {
+ _dcenter_x += fx / _mass * dt;
+ _dcenter_y += fy / _mass * dt;
+ _dtheta -= prod_vect(x - _center_x, y - _center_y, fx, fy) / (_mass * _moment_of_inertia) * dt;
+}
+
+void Polygon::apply_border_forces(scalar_t dt, scalar_t xmax, scalar_t ymax) {
+ for(int v = 0; v < _nb_vertices; v++) {
+ int vp = (v+1)%_nb_vertices;
+ for(int d = 0; d < _nb_dots[v]; d++) {
+ scalar_t s = scalar_t(d * dl)/_length[v];
+ scalar_t x = _x[v] * (1 - s) + _x[vp] * s;
+ scalar_t y = _y[v] * (1 - s) + _y[vp] * s;
+ scalar_t vx = 0, vy = 0;
+ if(x < 0) vx = x;
+ else if(x > xmax) vx = x - xmax;
+ if(y < 0) vy = y;
+ else if(y > ymax) vy = y - ymax;
+ apply_force(dt, x, y, - dl * vx * repulsion_constant, - dl * vy * repulsion_constant);
+ }
+ }
+}
+
+void Polygon::apply_collision_forces(scalar_t dt, int n_polygon, Polygon *p) {
+ scalar_t closest_x[_total_nb_dots], closest_y[_total_nb_dots];
+ bool inside[_total_nb_dots];
+ scalar_t distance[_total_nb_dots];
+ int _new_effecting_edge[_total_nb_dots];
+
+ int first_dot = 0;
+
+ for(int v = 0; v < _nb_vertices; v++) {
+ int vp = (v+1)%_nb_vertices;
+ scalar_t x = _x[v], y = _y[v], xp = _x[vp], yp = _y[vp];
+
+ for(int d = 0; d < _nb_dots[v]; d++) {
+ inside[d] = false;
+ distance[d] = FLT_MAX;
+ }
+
+ // First, we tag the dots located inside the polygon p
+
+ for(int t = 0; t < p->_nb_vertices-2; t++) {
+ scalar_t min = 0, max = 1;
+ scalar_t xa = p->_x[p->_triangles[t].a], ya = p->_y[p->_triangles[t].a];
+ scalar_t xb = p->_x[p->_triangles[t].b], yb = p->_y[p->_triangles[t].b];
+ scalar_t xc = p->_x[p->_triangles[t].c], yc = p->_y[p->_triangles[t].c];
+ scalar_t den, num;
+
+ const scalar_t eps = 1e-6;
+
+ den = prod_vect(xp - x, yp - y, xb - xa, yb - ya);
+ num = prod_vect(xa - x, ya - y, xb - xa, yb - ya);
+ if(den > eps) {
+ if(num / den < max) max = num / den;
+ } else if(den < -eps) {
+ if(num / den > min) min = num / den;
+ } else {
+ if(num < 0) { min = 1; max = 0; }
+ }
+
+ den = prod_vect(xp - x, yp - y, xc - xb, yc - yb);
+ num = prod_vect(xb - x, yb - y, xc - xb, yc - yb);
+ if(den > eps) {
+ if(num / den < max) max = num / den;
+ } else if(den < -eps) {
+ if(num / den > min) min = num / den;
+ } else {
+ if(num < 0) { min = 1; max = 0; }
+ }
+
+ den = prod_vect(xp - x, yp - y, xa - xc, ya - yc);
+ num = prod_vect(xc - x, yc - y, xa - xc, ya - yc);
+ if(den > eps) {
+ if(num / den < max) max = num / den;
+ } else if(den < -eps) {
+ if(num / den > min) min = num / den;
+ } else {
+ if(num < 0) { min = 1; max = 0; }
+ }
+
+ for(int d = 0; d < _nb_dots[v]; d++) {
+ scalar_t s = scalar_t(d * dl)/_length[v];
+ if(s >= min && s <= max) inside[d] = true;
+ }
+ }
+
+ // Then, we compute for each dot what is the closest point on
+ // the boundary of p
+
+ for(int m = 0; m < p->_nb_vertices; m++) {
+ int mp = (m+1)%p->_nb_vertices;
+ scalar_t xa = p->_x[m], ya = p->_y[m];
+ scalar_t xb = p->_x[mp], yb = p->_y[mp];
+ scalar_t gamma0 = ((x - xa) * (xb - xa) + (y - ya) * (yb - ya)) / sq(p->_length[m]);
+ scalar_t gamma1 = ((xp - x) * (xb - xa) + (yp - y) * (yb - ya)) / sq(p->_length[m]);
+ scalar_t delta0 = (prod_vect(xb - xa, yb - ya, x - xa, y - ya)) / p->_length[m];
+ scalar_t delta1 = (prod_vect(xb - xa, yb - ya, xp - x, yp - y)) / p->_length[m];
+
+ for(int d = 0; d < _nb_dots[v]; d++) if(inside[d]) {
+ int r = _effecting_edge[(first_dot + d) * _nb_polygons + n_polygon];
+
+ // If there is already a spring, we look only at the
+ // vertices next to the current one
+
+ if(r < 0 || m == r || m == (r+1)%p->_nb_vertices || (m+1)%p->_nb_vertices == r) {
+
+ scalar_t s = scalar_t(d * dl)/_length[v];
+ scalar_t delta = abs(s * delta1 + delta0);
+ if(delta < distance[d]) {
+ scalar_t gamma = s * gamma1 + gamma0;
+ if(gamma < 0) {
+ scalar_t l = sqrt(sq(x * (1 - s) + xp * s - xa) + sq(y * (1 - s) + yp * s - ya));
+ if(l < distance[d]) {
+ distance[d] = l;
+ closest_x[d] = xa;
+ closest_y[d] = ya;
+ _new_effecting_edge[first_dot + d] = m;
+ }
+ } else if(gamma > 1) {
+ scalar_t l = sqrt(sq(x * (1 - s) + xp * s - xb) + sq(y * (1 - s) + yp * s - yb));
+ if(l < distance[d]) {
+ distance[d] = l;
+ closest_x[d] = xb;
+ closest_y[d] = yb;
+ _new_effecting_edge[first_dot + d] = m;
+ }
+ } else {
+ distance[d] = delta;
+ closest_x[d] = xa * (1 - gamma) + xb * gamma;
+ closest_y[d] = ya * (1 - gamma) + yb * gamma;
+ _new_effecting_edge[first_dot + d] = m;
+ }
+ }
+ }
+ } else _new_effecting_edge[first_dot + d] = -1;
+ }
+
+ // Apply forces
+
+ for(int d = 0; d < _nb_dots[v]; d++) if(inside[d]) {
+ scalar_t s = scalar_t(d * dl)/_length[v];
+ scalar_t x = _x[v] * (1 - s) + _x[vp] * s;
+ scalar_t y = _y[v] * (1 - s) + _y[vp] * s;
+ scalar_t vx = x - closest_x[d];
+ scalar_t vy = y - closest_y[d];
+
+ p->apply_force(dt,
+ closest_x[d], closest_y[d],
+ dl * vx * repulsion_constant, dl * vy * repulsion_constant);
+
+ apply_force(dt,
+ x, y,
+ - dl * vx * repulsion_constant, - dl * vy * repulsion_constant);
+ }
+
+ first_dot += _nb_dots[v];
+ }
+
+ for(int d = 0; d < _total_nb_dots; d++)
+ _effecting_edge[d * _nb_polygons + n_polygon] = _new_effecting_edge[d];
+
+}
+
+bool Polygon::collide(Polygon *p) {
+ for(int n = 0; n < _nb_vertices; n++) {
+ int np = (n+1)%_nb_vertices;
+ for(int m = 0; m < p->_nb_vertices; m++) {
+ int mp = (m+1)%p->_nb_vertices;
+ scalar_t det, s = -1, t = -1;
+ intersection(_x[n], _y[n], _x[np], _y[np],
+ p->_x[m], p->_y[m], p->_x[mp], p->_y[mp], det, s, t);
+ if(det != 0 && s>= 0 && s <= 1&& t >= 0 && t <= 1) return true;
+ }
+ }
+
+ for(int n = 0; n < _nb_vertices; n++) if(p->contain(_x[n], _y[n])) return true;
+ for(int n = 0; n < p->_nb_vertices; n++) if(contain(p->_x[n], p->_y[n])) return true;
+
+ return false;
+}
--- /dev/null
+
+/*
+ * dyncnn is a deep-learning algorithm for the prediction of
+ * interacting object dynamics
+ *
+ * Copyright (c) 2016 Idiap Research Institute, http://www.idiap.ch/
+ * Written by Francois Fleuret <francois.fleuret@idiap.ch>
+ *
+ * This file is part of dyncnn.
+ *
+ * dyncnn is free software: you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 3 as
+ * published by the Free Software Foundation.
+ *
+ * dyncnn is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with dyncnn. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#ifndef POLYGON_H
+#define POLYGON_H
+
+#include "misc.h"
+#include "canvas.h"
+
+#ifdef XFIG_SUPPORT
+#include "xfig_tracer.h"
+#endif
+
+#ifdef X11_SUPPORT
+#include "simple_window.h"
+#endif
+
+class Polygon {
+ struct Triangle {
+ int a, b, c;
+ };
+
+ int _nb_max_polygons;
+ int _nb_polygons; // We need to know the total number of polygons in
+ // the universe
+
+ scalar_t _mass, _moment_of_inertia, _radius;
+
+ scalar_t *_relative_x, *_relative_y;
+ scalar_t _last_center_x, _last_center_y, _last_theta;
+
+ int _total_nb_dots;
+ int *_nb_dots;
+ int *_effecting_edge;
+
+ scalar_t *_length;
+ Triangle *_triangles;
+
+ inline void intersection(scalar_t xa2, scalar_t ya2,
+ scalar_t xa1, scalar_t ya1,
+ scalar_t xb2, scalar_t yb2,
+ scalar_t xb1, scalar_t yb1,
+ scalar_t &det, scalar_t &s1, scalar_t &s2) {
+ scalar_t m11, m12, m21, m22, n11, n12, n21, n22, v1, v2;
+ m11 = xa1 - xa2; m12 = xb2 - xb1; m21 = ya1 - ya2; m22 = yb2 - yb1;
+ det = m11 * m22 - m12 * m21;
+ if(det != 0) {
+ n11 = + m22 / det; n12 = - m12 / det; n21 = - m21 / det; n22 = + m11 / det;
+ v1 = xb2 - xa2; v2 = yb2 - ya2;
+ s1 = n11 * v1 + n12 * v2; s2 = n21 * v1 + n22 * v2;
+ }
+ }
+
+public:
+ bool _initialized;
+ bool _nailed;
+ int _nb_vertices;
+ scalar_t *_x, *_y;
+ scalar_t _red, _green, _blue;
+ scalar_t _center_x, _center_y, _theta;
+ scalar_t _dcenter_x, _dcenter_y, _dtheta;
+
+ // The x and y parameters are ignored if null. Useful to initialize
+ // directly.
+
+ Polygon(scalar_t mass,
+ scalar_t red, scalar_t green, scalar_t blue,
+ scalar_t *x, scalar_t *y,
+ int nv);
+
+ ~Polygon();
+
+ Polygon *clone();
+
+#ifdef XFIG_SUPPORT
+ void color_xfig(XFigTracer *tracer);
+ void print_xfig(XFigTracer *tracer);
+#endif
+
+#ifdef X11_SUPPORT
+ void draw(SimpleWindow *window);
+ void draw_contours(SimpleWindow *window);
+#endif
+
+ void draw(Canvas *canvas);
+ void draw_contours(Canvas *canvas);
+
+ void set_vertex(int k, scalar_t x, scalar_t y);
+ void set_position(scalar_t center_x, scalar_t center_y, scalar_t theta);
+ void set_speed(scalar_t dcenter_x, scalar_t dcenter_y, scalar_t dtheta);
+ bool contain(scalar_t x, scalar_t y);
+ void triangularize(int &nt, int nb, int *index);
+ void initialize(int nb_polygons);
+ bool update(scalar_t dt);
+ scalar_t relative_x(scalar_t ax, scalar_t ay);
+ scalar_t relative_y(scalar_t ax, scalar_t ay);
+ scalar_t absolute_x(scalar_t rx, scalar_t ry);
+ scalar_t absolute_y(scalar_t rx, scalar_t ry);
+
+ void apply_force(scalar_t dt, scalar_t x, scalar_t y, scalar_t fx, scalar_t fy);
+ void apply_border_forces(scalar_t dt, scalar_t xmax, scalar_t ymax);
+ void apply_collision_forces(scalar_t dt, int n_polygon, Polygon *p);
+
+ bool collide(Polygon *p);
+};
+
+#endif
--- /dev/null
+#!/usr/bin/env python-for-pytorch
+
+import torch
+import torchvision
+from torchvision import datasets
+
+from _ext import mylib
+
+x = torch.ByteTensor(4, 5).fill_(0)
+
+print(x.size())
+
+mylib.generate_sequence(x)
+
+print(x.size())
+
+x = x.float().sub_(128).div_(128)
+
+torchvision.utils.save_image(x[0], 'example.png')
--- /dev/null
+
+/*
+ * dyncnn is a deep-learning algorithm for the prediction of
+ * interacting object dynamics
+ *
+ * Copyright (c) 2016 Idiap Research Institute, http://www.idiap.ch/
+ * Written by Francois Fleuret <francois.fleuret@idiap.ch>
+ *
+ * This file is part of dyncnn.
+ *
+ * dyncnn is free software: you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 3 as
+ * published by the Free Software Foundation.
+ *
+ * dyncnn is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with dyncnn. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#include <string.h>
+
+#include "universe.h"
+
+Universe::Universe(int nb_max_polygons,
+ scalar_t width, scalar_t height) : _width(width),
+ _height(height),
+ _nb_max_polygons(nb_max_polygons),
+ _nb_polygons(0) {
+ _polygons = new Polygon *[_nb_max_polygons];
+ for(int n = 0; n < _nb_max_polygons; n++) _polygons[n] = 0;
+}
+
+Universe::~Universe() {
+ clear();
+ delete[] _polygons;
+}
+
+void Universe::initialize_polygon(Polygon *p) {
+ p->initialize(_nb_max_polygons);
+}
+
+void Universe::clear() {
+ for(int n = 0; n < _nb_polygons; n++) if(_polygons[n]) delete _polygons[n];
+ _nb_polygons = 0;
+}
+
+void Universe::add_polygon(Polygon *p) {
+ if(_nb_polygons < _nb_max_polygons) {
+ if(!p->_initialized) {
+ cerr << "You can not add a non-initialized polygon." << endl;
+ abort();
+ }
+ _polygons[_nb_polygons++] = p;
+ } else {
+ cerr << "Too many polygons!" << endl;
+ exit(1);
+ }
+}
+
+bool Universe::collide(Polygon *p) {
+ for(int n = 0; n < _nb_polygons; n++)
+ if(_polygons[n] && _polygons[n]->collide(p)) return true;
+
+ return false;
+}
+
+void Universe::compute_pseudo_collisions(int nb_axis, int *nb_colliding_axis) {
+ Couple couples[_nb_polygons * 2];
+ int in[_nb_polygons];
+ memset((void *) nb_colliding_axis, 0, _nb_polygons * _nb_polygons * sizeof(int));
+
+ for(int a = 0; a < nb_axis; a++) {
+ scalar_t alpha = M_PI * scalar_t(a) / scalar_t(nb_axis);
+ scalar_t vx = cos(alpha), vy = sin(alpha);
+
+ for(int n = 0; n < _nb_polygons; n++) {
+ scalar_t *x = _polygons[n]->_x, *y = _polygons[n]->_y;
+ scalar_t min = x[0] * vx + y[0] * vy, max = min;
+
+ for(int v = 1; v < _polygons[n]->_nb_vertices; v++) {
+ scalar_t s = x[v] * vx + y[v] * vy;
+ if(s > max) max = s;
+ if(s < min) min = s;
+ }
+
+ couples[2 * n + 0].value = min;
+ couples[2 * n + 0].index = n;
+ couples[2 * n + 1].value = max;
+ couples[2 * n + 1].index = n;
+ }
+
+ qsort((void *) couples, 2 * _nb_polygons, sizeof(Couple), compare_couple);
+
+ int nb_in = 0;
+ memset((void *) in, 0, _nb_polygons * sizeof(int));
+ for(int k = 0; k < 2 * _nb_polygons; k++) {
+ int i = couples[k].index;
+ in[i] = !in[i];
+ if(in[i]) {
+ nb_in++;
+ if(nb_in > 1) {
+ for(int j = 0; j < i; j++)
+ if(j != i && in[j]) nb_colliding_axis[j + i * _nb_polygons]++;
+ for(int j = i+1; j < _nb_polygons; j++)
+ if(j != i && in[j]) nb_colliding_axis[i + j * _nb_polygons]++;
+ }
+ } else nb_in--;
+ }
+ }
+
+ for(int i = 0; i < _nb_polygons; i++) {
+ for(int j = 0; j < i; j++) {
+ if(nb_colliding_axis[j + i * _nb_polygons] > nb_colliding_axis[i + i * _nb_polygons])
+ nb_colliding_axis[i + i * _nb_polygons] = nb_colliding_axis[j + i * _nb_polygons];
+ nb_colliding_axis[i + j * _nb_polygons] = nb_colliding_axis[j + i * _nb_polygons];
+ }
+ }
+}
+
+bool Universe::update(scalar_t dt) {
+ bool result = false;
+ apply_collision_forces(dt);
+ for(int n = 0; n < _nb_polygons; n++) if(_polygons[n]) {
+ _polygons[n]->apply_border_forces(dt, _width, _height);
+ result |= _polygons[n]->update(dt);
+ }
+ return result;
+}
+
+Polygon *Universe::pick_polygon(scalar_t x, scalar_t y) {
+ for(int n = 0; n < _nb_polygons; n++)
+ if(_polygons[n] && _polygons[n]->contain(x, y)) return _polygons[n];
+ return 0;
+}
+
+void Universe::draw(Canvas *canvas) {
+ for(int n = 0; n < _nb_polygons; n++) {
+ if(_polygons[n]) {
+ _polygons[n]->draw(canvas);
+ }
+ }
+
+ for(int n = 0; n < _nb_polygons; n++) {
+ if(_polygons[n]) {
+ _polygons[n]->draw_contours(canvas);
+ }
+ }
+}
+
+void Universe::apply_collision_forces(scalar_t dt) {
+ const int nb_axis = 2;
+ int nb_collision[_nb_polygons * _nb_polygons];
+
+ compute_pseudo_collisions(nb_axis, nb_collision);
+
+ for(int n = 0; n < _nb_polygons; n++) if(_polygons[n])
+ for(int m = 0; m < _nb_polygons; m++)
+ if(m != n && _polygons[m] && nb_collision[n + _nb_polygons * m] == nb_axis)
+ _polygons[n]->apply_collision_forces(dt, m, _polygons[m]);
+}
--- /dev/null
+
+/*
+ * dyncnn is a deep-learning algorithm for the prediction of
+ * interacting object dynamics
+ *
+ * Copyright (c) 2016 Idiap Research Institute, http://www.idiap.ch/
+ * Written by Francois Fleuret <francois.fleuret@idiap.ch>
+ *
+ * This file is part of dyncnn.
+ *
+ * dyncnn is free software: you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 3 as
+ * published by the Free Software Foundation.
+ *
+ * dyncnn is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with dyncnn. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#ifndef UNIVERSE_H
+#define UNIVERSE_H
+
+#include <iostream>
+#include <cmath>
+
+#include "misc.h"
+#include "canvas.h"
+#include "polygon.h"
+
+using namespace std;
+
+class Universe {
+ scalar_t _width, _height;
+public:
+ int _nb_max_polygons, _nb_polygons;
+ Polygon **_polygons;
+
+ inline scalar_t width() { return _width; }
+ inline scalar_t height() { return _height; }
+
+ Universe(int nb_max_polygons, scalar_t width, scalar_t height);
+ // The destructor deletes all the added polygons
+ ~Universe();
+
+ void initialize_polygon(Polygon *p);
+ void clear();
+ void add_polygon(Polygon *p);
+ bool collide(Polygon *p);
+
+ // Compute collisions between projections of the polygons on a few
+ // axis to speed up the computation
+ void compute_pseudo_collisions(int nb_axis, int *nb_colliding_axis);
+ void apply_collision_forces(scalar_t dt);
+ bool update(scalar_t dt);
+
+ Polygon *pick_polygon(scalar_t x, scalar_t y);
+
+ void draw(Canvas *canvas);
+};
+
+#endif