X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=world.py;h=118a470b29b159d97826dc895362db8e8d673ded;hb=8adf0586ee5aeb9fbdf81b78c7ff4b484a9b82ab;hp=fb5d5c796990ef476ff79fb9ead66615a17fd7fd;hpb=a8f039a9b491b1b4b47f6b9f8123c7261e758661;p=culture.git

diff --git a/world.py b/world.py
index fb5d5c7..118a470 100755
--- a/world.py
+++ b/world.py
@@ -1,259 +1,129 @@
 #!/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),
-    ]
-
-    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
-
-        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))
-
-        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
+colors = torch.tensor(
+    [
+        [255, 255, 255],
+        [0, 0, 0],
+        [255, 0, 0],
+        [0, 128, 0],
+        [0, 0, 255],
+        [255, 200, 0],
+        [192, 192, 192],
+    ]
+)
+
+token2char = "_X" + "".join([str(n) for n in range(len(colors) - 2)]) + ">"
+
+
+def generate(
+    nb,
+    height,
+    width,
+    max_nb_obj=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))
+
+    for n in tqdm.tqdm(range(nb), dynamic_ncols=True, desc="world generation"):
+        nb_fish = torch.randint(max_nb_obj, (1,)).item() + 1
+        for c in torch.randperm(colors.size(0) - 2)[:nb_fish].sort().values:
+            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, max_nb_obj=3)
+    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)