#!/usr/bin/env python
-import math
+# Any copyright is dedicated to the Public Domain.
+# https://creativecommons.org/publicdomain/zero/1.0/
+
+# Written by Francois Fleuret <francois@fleuret.org>
+
+import math, sys, tqdm
import torch, torchvision
from torch import nn
from torch.nn import functional as F
-import cairo
-
-
-class Box:
- def __init__(self, x, y, w, h, r, g, b):
- self.x = x
- self.y = y
- self.w = w
- self.h = h
- self.r = r
- self.g = g
- self.b = b
-
- def collision(self, scene):
- for c in scene:
- if (
- self is not c
- and max(self.x, c.x) <= min(self.x + self.w, c.x + c.w)
- and max(self.y, c.y) <= min(self.y + self.h, c.y + c.h)
- ):
- return True
- return False
-
-
-def scene2tensor(xh, yh, scene, size=64):
- width, height = size, size
- pixel_map = torch.ByteTensor(width, height, 4).fill_(255)
- data = pixel_map.numpy()
- surface = cairo.ImageSurface.create_for_data(
- data, cairo.FORMAT_ARGB32, width, height
- )
-
- ctx = cairo.Context(surface)
- ctx.set_fill_rule(cairo.FILL_RULE_EVEN_ODD)
-
- for b in scene:
- ctx.move_to(b.x * size, b.y * size)
- ctx.rel_line_to(b.w * size, 0)
- ctx.rel_line_to(0, b.h * size)
- ctx.rel_line_to(-b.w * size, 0)
- ctx.close_path()
- ctx.set_source_rgba(b.r, b.g, b.b, 1.0)
- ctx.fill()
- hs = size * 0.1
- ctx.set_source_rgba(0.0, 0.0, 0.0, 1.0)
- ctx.move_to(xh * size - hs / 2, yh * size - hs / 2)
- ctx.rel_line_to(hs, 0)
- ctx.rel_line_to(0, hs)
- ctx.rel_line_to(-hs, 0)
- ctx.close_path()
- ctx.fill()
-
- return pixel_map[None, :, :, :3].flip(-1).permute(0, 3, 1, 2).float() / 255
-
-
-def random_scene():
- scene = []
- colors = [
- (1.00, 0.00, 0.00),
- (0.00, 1.00, 0.00),
- (0.60, 0.60, 1.00),
- (1.00, 1.00, 0.00),
- (0.75, 0.75, 0.75),
- ]
-
- for k in range(10):
- wh = torch.rand(2) * 0.2 + 0.2
- xy = torch.rand(2) * (1 - wh)
- c = colors[torch.randint(len(colors), (1,))]
- b = Box(
- xy[0].item(), xy[1].item(), wh[0].item(), wh[1].item(), c[0], c[1], c[2]
- )
- if not b.collision(scene):
- scene.append(b)
-
- return scene
+######################################################################
-def sequence(nb_steps=10, all_frames=False):
- delta = 0.1
- effects = [
- (False, 0, 0),
- (False, delta, 0),
- (False, 0, delta),
- (False, -delta, 0),
- (False, 0, -delta),
- (True, delta, 0),
- (True, 0, delta),
- (True, -delta, 0),
- (True, 0, -delta),
+colors = torch.tensor(
+ [
+ [255, 255, 255],
+ [0, 0, 0],
+ [255, 0, 0],
+ [0, 128, 0],
+ [0, 0, 255],
+ [255, 255, 0],
+ [192, 192, 192],
]
+)
- while True:
- frames = []
-
- scene = random_scene()
- xh, yh = tuple(x.item() for x in torch.rand(2))
-
- frames.append(scene2tensor(xh, yh, scene))
-
- actions = torch.randint(len(effects), (nb_steps,))
- change = False
+token2char = "_X" + "".join([str(n) for n in range(len(colors) - 2)]) + ">"
- for a in actions:
- g, dx, dy = effects[a]
- if g:
- for b in scene:
- if b.x <= xh and b.x + b.w >= xh and b.y <= yh and b.y + b.h >= yh:
- x, y = b.x, b.y
- b.x += dx
- b.y += dy
- if (
- b.x < 0
- or b.y < 0
- or b.x + b.w > 1
- or b.y + b.h > 1
- or b.collision(scene)
- ):
- b.x, b.y = x, y
- else:
- xh += dx
- yh += dy
- change = True
- else:
- x, y = xh, yh
- xh += dx
- yh += dy
- if xh < 0 or xh > 1 or yh < 0 or yh > 1:
- xh, yh = x, y
- if all_frames:
- frames.append(scene2tensor(xh, yh, scene))
+def generate(
+ nb,
+ height,
+ width,
+ max_nb_obj=len(colors) - 2,
+ nb_iterations=2,
+):
+ f_start = torch.zeros(nb, height, width, dtype=torch.int64)
+ f_end = torch.zeros(nb, height, width, dtype=torch.int64)
+ n = torch.arange(f_start.size(0))
- if not all_frames:
- frames.append(scene2tensor(xh, yh, scene))
-
- if change:
- break
-
- return frames, actions
-
-
-######################################################################
-
-
-# ||x_i - c_j||^2 = ||x_i||^2 + ||c_j||^2 - 2<x_i, c_j>
-def sq2matrix(x, c):
- nx = x.pow(2).sum(1)
- nc = c.pow(2).sum(1)
- return nx[:, None] + nc[None, :] - 2 * x @ c.t()
-
-
-def update_centroids(x, c, nb_min=1):
- _, b = sq2matrix(x, c).min(1)
- b.squeeze_()
- nb_resets = 0
+ for n in range(nb):
+ nb_fish = torch.randint(max_nb_obj, (1,)).item() + 1
+ for c in range(nb_fish):
+ i, j = (
+ torch.randint(height - 2, (1,))[0] + 1,
+ torch.randint(width - 2, (1,))[0] + 1,
+ )
+ vm = torch.randint(4, (1,))[0]
+ vi, vj = (vm // 2) * (2 * (vm % 2) - 1), (1 - vm // 2) * (2 * (vm % 2) - 1)
+
+ f_start[n, i, j] = c + 2
+ f_start[n, i - vi, j - vj] = c + 2
+ f_start[n, i + vj, j - vi] = c + 2
+ f_start[n, i - vj, j + vi] = c + 2
+
+ for l in range(nb_iterations):
+ i += vi
+ j += vj
+ if i < 0 or i >= height or j < 0 or j >= width:
+ i -= vi
+ j -= vj
+ vi, vj = -vi, -vj
+ i += vi
+ j += vj
+
+ f_end[n, i, j] = c + 2
+ f_end[n, i - vi, j - vj] = c + 2
+ f_end[n, i + vj, j - vi] = c + 2
+ f_end[n, i - vj, j + vi] = c + 2
+
+ return torch.cat(
+ [
+ f_end.flatten(1),
+ torch.full((f_end.size(0), 1), len(colors)),
+ f_start.flatten(1),
+ ],
+ dim=1,
+ )
- for k in range(0, c.size(0)):
- i = b.eq(k).nonzero(as_tuple=False).squeeze()
- if i.numel() >= nb_min:
- c[k] = x.index_select(0, i).mean(0)
- else:
- n = torch.randint(x.size(0), (1,))
- nb_resets += 1
- c[k] = x[n]
- return c, b, nb_resets
+def sample2img(seq, height, width):
+ f_start = seq[:, : height * width].reshape(-1, height, width)
+ f_start = (f_start >= len(colors)).long() + (f_start < len(colors)).long() * f_start
+ f_end = seq[:, height * width + 1 :].reshape(-1, height, width)
+ f_end = (f_end >= len(colors)).long() + (f_end < len(colors)).long() * f_end
+ img_f_start, img_f_end = colors[f_start], colors[f_end]
-def kmeans(x, nb_centroids, nb_min=1):
- if x.size(0) < nb_centroids * nb_min:
- print("Not enough points!")
- exit(1)
+ img = torch.cat(
+ [
+ img_f_start,
+ torch.full(
+ (img_f_start.size(0), img_f_start.size(1), 1, img_f_start.size(3)), 1
+ ),
+ img_f_end,
+ ],
+ dim=2,
+ )
- c = x[torch.randperm(x.size(0))[:nb_centroids]]
- t = torch.full((x.size(0),), -1)
- n = 0
+ return img.permute(0, 3, 1, 2)
- while True:
- c, u, nb_resets = update_centroids(x, c, nb_min)
- n = n + 1
- nb_changes = (u - t).sign().abs().sum() + nb_resets
- t = u
- if nb_changes == 0:
- break
- return c, t
+def seq2str(seq):
+ result = []
+ for s in seq:
+ result.append("".join([token2char[v] for v in s]))
+ return result
######################################################################
-
-def patchify(x, factor, invert_size=None):
- if invert_size is None:
- return (
- x.reshape(
- x.size(0), #0
- x.size(1), #1
- factor, #2
- x.size(2) // factor,#3
- factor,#4
- x.size(3) // factor,#5
- )
- .permute(0, 2, 4, 1, 3, 5)
- .reshape(-1, x.size(1), x.size(2) // factor, x.size(3) // factor)
- )
- else:
- return (
- x.reshape(
- invert_size[0], #0
- factor, #1
- factor, #2
- invert_size[1], #3
- invert_size[2] // factor, #4
- invert_size[3] // factor, #5
- )
- .permute(0, 3, 1, 4, 2, 5)
- .reshape(invert_size)
- )
-
-
if __name__ == "__main__":
import time
- all_frames = []
- nb = 1000
+ height, width = 6, 8
start_time = time.perf_counter()
- for n in range(nb):
- frames, actions = sequence(nb_steps=31)
- all_frames += frames
- end_time = time.perf_counter()
- print(f"{nb / (end_time - start_time):.02f} samples per second")
-
- input = torch.cat(all_frames, 0)
- x = patchify(input, 8)
- y = x.reshape(x.size(0), -1)
- print(f"{x.size()=} {y.size()=}")
- centroids, t = kmeans(y, 4096)
- results = centroids[t]
- results = results.reshape(x.size())
- results = patchify(results, 8, input.size())
+ seq = generate(nb=64, height=height, width=width)
+ delay = time.perf_counter() - start_time
+ print(f"{seq.size(0)/delay:02f} samples/s")
- print(f"{input.size()=} {results.size()=}")
+ print(seq2str(seq[:4]))
- torchvision.utils.save_image(input[:64], "orig.png", nrow=8)
- torchvision.utils.save_image(results[:64], "qtiz.png", nrow=8)
+ img = sample2img(seq, height, width)
+ print(img.size())
- # frames, actions = sequence(nb_steps=31, all_frames=True)
- # frames = torch.cat(frames, 0)
- # torchvision.utils.save_image(frames, "seq.png", nrow=8)
+ torchvision.utils.save_image(img.float() / 255.0, "world.png", nrow=8, padding=2)