sky.py

   1 #!/usr/bin/env python
   2
   3 # Any copyright is dedicated to the Public Domain.
   4 # https://creativecommons.org/publicdomain/zero/1.0/
   5
   6 # Written by Francois Fleuret <francois@fleuret.org>
   7
   8 import math, sys, tqdm, os
   9
  10 import torch, torchvision
  11
  12 from torch import nn
  13 from torch.nn import functional as F
  14
  15 ######################################################################
  16
  17 import problem
  18
  19
  20 class Sky(problem.Problem):
  21     colors = torch.tensor(
  22         [
  23             [255, 255, 255],
  24             [255, 0, 0],
  25             [0, 192, 0],
  26             [0, 0, 255],
  27             [255, 192, 0],
  28             [0, 255, 255],
  29             [255, 0, 255],
  30             [192, 255, 192],
  31             [255, 192, 192],
  32             [192, 192, 255],
  33             [192, 192, 192],
  34         ]
  35     )
  36
  37     token_background = 0
  38     first_bird_token = 1
  39     nb_bird_tokens = colors.size(0) - 1
  40     token_forward = first_bird_token + nb_bird_tokens
  41     token_backward = token_forward + 1
  42
  43     token2char = (
  44         "_" + "".join([chr(ord("A") + n) for n in range(len(colors) - 1)]) + "><"
  45     )
  46
  47     def __init__(
  48         self,
  49         height=6,
  50         width=8,
  51         nb_birds=3,
  52         speed=2,
  53         nb_iterations=2,
  54         avoid_collision=True,
  55     ):
  56         self.height = height
  57         self.width = width
  58         self.nb_birds = nb_birds
  59         self.speed = speed
  60         self.nb_iterations = nb_iterations
  61         self.avoid_collision = avoid_collision
  62
  63     def direction_tokens(self):
  64         return self.token_forward, self.token_backward
  65
  66     def generate_frame_sequences(self, nb):
  67         frame_sequences = []
  68
  69         for _ in tqdm.tqdm(range(nb), dynamic_ncols=True, desc="world generation"):
  70             i, j, vi, vj = (
  71                 torch.empty(self.nb_birds, dtype=torch.int64),
  72                 torch.empty(self.nb_birds, dtype=torch.int64),
  73                 torch.empty(self.nb_birds, dtype=torch.int64),
  74                 torch.empty(self.nb_birds, dtype=torch.int64),
  75             )
  76
  77             def collision_okay():
  78                 if not self.avoid_collision:
  79                     return True
  80
  81                 count = torch.zeros(self.height, self.width, dtype=torch.int64)
  82
  83                 for n in range(self.nb_birds):
  84                     count[i[n], j[n]] += 1
  85                     count[i[n] - vi[n], j[n]] += 1
  86                     count[i[n], j[n] - vj[n]] += 1
  87
  88                 return count.max() <= 1
  89
  90             col = (
  91                 torch.randperm(self.colors.size(0) - 1)[: self.nb_birds].sort().values
  92                 + 1
  93             )
  94
  95             while True:
  96                 while True:
  97                     for n in range(self.nb_birds):
  98                         while True:
  99                             i[n] = torch.randint(self.height, (1,))
 100                             j[n] = torch.randint(self.width, (1,))
 101                             vm = torch.randint(4, (1,))
 102                             vi[n], vj[n] = (vm % 2) * 2 - 1, (vm // 2) * 2 - 1
 103                             if (
 104                                 i[n] - vi[n] >= 0
 105                                 and i[n] - vi[n] < self.height
 106                                 and j[n] - vj[n] >= 0
 107                                 and j[n] - vj[n] < self.width
 108                             ):
 109                                 break
 110
 111                     if collision_okay():
 112                         break
 113
 114                 result = torch.zeros(
 115                     self.nb_iterations * self.speed,
 116                     self.height,
 117                     self.width,
 118                     dtype=torch.int64,
 119                 )
 120
 121                 for l in range(self.nb_iterations * self.speed):
 122                     fine = collision_okay()
 123                     for n in range(self.nb_birds):
 124                         c = col[n]
 125                         result[l, i[n], j[n]] = c
 126                         result[l, i[n] - vi[n], j[n]] = c
 127                         result[l, i[n], j[n] - vj[n]] = c
 128
 129                         if (i[n] == 0 and vi[n] == -1) or (
 130                             i[n] == self.height - 1 and vi[n] == 1
 131                         ):
 132                             vi[n] = -vi[n]
 133
 134                         if (j[n] == 0 and vj[n] == -1) or (
 135                             j[n] == self.width - 1 and vj[n] == 1
 136                         ):
 137                             vj[n] = -vj[n]
 138
 139                         i[n] += vi[n]
 140                         j[n] += vj[n]
 141
 142                 if fine:
 143                     break
 144
 145             frame_sequences.append(
 146                 result[
 147                     torch.arange(self.nb_iterations, device=result.device) * self.speed
 148                 ]
 149             )
 150
 151         return frame_sequences
 152
 153     ######################################################################
 154
 155     def generate_token_sequences(self, nb):
 156         frame_sequences = self.generate_frame_sequences(nb)
 157
 158         result = []
 159
 160         for frame_sequence in frame_sequences:
 161             a = []
 162             if torch.rand(1) < 0.5:
 163                 for frame in frame_sequence:
 164                     if len(a) > 0:
 165                         a.append(torch.tensor([self.token_forward]))
 166                     a.append(frame.flatten())
 167             else:
 168                 for frame in reversed(frame_sequence):
 169                     if len(a) > 0:
 170                         a.append(torch.tensor([self.token_backward]))
 171                     a.append(frame.flatten())
 172
 173             result.append(torch.cat(a, dim=0)[None, :])
 174
 175         return torch.cat(result, dim=0)
 176
 177     ######################################################################
 178
 179     def frame2img(self, x, scale=15):
 180         x = x.reshape(-1, self.height, self.width)
 181         m = torch.logical_and(
 182             x >= 0, x < self.first_bird_token + self.nb_bird_tokens
 183         ).long()
 184         x = self.colors[x * m].permute(0, 3, 1, 2)
 185         s = x.shape
 186         x = x[:, :, :, None, :, None].expand(-1, -1, -1, scale, -1, scale)
 187         x = x.reshape(s[0], s[1], s[2] * scale, s[3] * scale)
 188
 189         x[:, :, :, torch.arange(0, x.size(3), scale)] = 0
 190         x[:, :, torch.arange(0, x.size(2), scale), :] = 0
 191         x = x[:, :, 1:, 1:]
 192
 193         for n in range(m.size(0)):
 194             for i in range(m.size(1)):
 195                 for j in range(m.size(2)):
 196                     if m[n, i, j] == 0:
 197                         for k in range(2, scale - 2):
 198                             for l in [0, 1]:
 199                                 x[n, :, i * scale + k, j * scale + k - l] = 0
 200                                 x[
 201                                     n, :, i * scale + scale - 1 - k, j * scale + k - l
 202                                 ] = 0
 203
 204         return x
 205
 206     def seq2img(self, seq, scale=15):
 207         all = [
 208             self.frame2img(
 209                 seq[:, : self.height * self.width].reshape(-1, self.height, self.width),
 210                 scale,
 211             )
 212         ]
 213
 214         separator = torch.full((seq.size(0), 3, self.height * scale - 1, 1), 0)
 215
 216         t = self.height * self.width
 217
 218         while t < seq.size(1):
 219             direction_tokens = seq[:, t]
 220             t += 1
 221
 222             direction_images = self.colors[
 223                 torch.full(
 224                     (direction_tokens.size(0), self.height * scale - 1, scale), 0
 225                 )
 226             ].permute(0, 3, 1, 2)
 227
 228             for n in range(direction_tokens.size(0)):
 229                 if direction_tokens[n] == self.token_forward:
 230                     for k in range(scale):
 231                         for l in [0, 1]:
 232                             direction_images[
 233                                 n,
 234                                 :,
 235                                 (self.height * scale) // 2 - scale // 2 + k - l,
 236                                 3 + scale // 2 - abs(k - scale // 2),
 237                             ] = 0
 238                 elif direction_tokens[n] == self.token_backward:
 239                     for k in range(scale):
 240                         for l in [0, 1]:
 241                             direction_images[
 242                                 n,
 243                                 :,
 244                                 (self.height * scale) // 2 - scale // 2 + k - l,
 245                                 3 + abs(k - scale // 2),
 246                             ] = 0
 247                 else:
 248                     for k in range(2, scale - 2):
 249                         for l in [0, 1]:
 250                             direction_images[
 251                                 n,
 252                                 :,
 253                                 (self.height * scale) // 2 - scale // 2 + k - l,
 254                                 k,
 255                             ] = 0
 256                             direction_images[
 257                                 n,
 258                                 :,
 259                                 (self.height * scale) // 2 - scale // 2 + k - l,
 260                                 scale - 1 - k,
 261                             ] = 0
 262
 263             all += [
 264                 separator,
 265                 direction_images,
 266                 separator,
 267                 self.frame2img(
 268                     seq[:, t : t + self.height * self.width].reshape(
 269                         -1, self.height, self.width
 270                     ),
 271                     scale,
 272                 ),
 273             ]
 274
 275             t += self.height * self.width
 276
 277         return torch.cat(all, dim=3)
 278
 279     def seq2str(self, seq):
 280         result = []
 281         for s in seq:
 282             result.append("".join([self.token2char[v] for v in s]))
 283         return result
 284
 285     def save_image(self, input, result_dir, filename):
 286         img = self.seq2img(input.to("cpu"))
 287         image_name = os.path.join(result_dir, filename)
 288         torchvision.utils.save_image(img.float() / 255.0, image_name, nrow=6, padding=4)
 289
 290     def save_quizzes(self, input, result_dir, filename_prefix):
 291         self.save_image(input, result_dir, filename_prefix + ".png")
 292
 293
 294 ######################################################################
 295
 296 if __name__ == "__main__":
 297     import time
 298
 299     sky = Sky(height=6, width=8, speed=2, nb_iterations=2)
 300
 301     start_time = time.perf_counter()
 302     token_sequences = sky.generate_token_sequences(nb=64)
 303     delay = time.perf_counter() - start_time
 304     print(f"{token_sequences.size(0)/delay:02f} seq/s")
 305
 306     # print(sky.seq2str(seq[:4]))
 307
 308     # for t in range(len(it[0])):
 309     # img = torch.cat([sky.frame2img(f[t]) for f in it], dim=0)
 310     # torchvision.utils.save_image(
 311     # img.float() / 255.0,
 312     # f"/tmp/frame_{t:03d}.png",
 313     # nrow=8,
 314     # padding=6,
 315     # pad_value=0,
 316     # )
 317
 318     # m = (torch.rand(seq.size()) < 0.05).long()
 319     # seq = (1 - m) * seq + m * 23
 320
 321     # print(seq.size())
 322     img = sky.seq2img(token_sequences)
 323     # print(img.size())
 324
 325     torchvision.utils.save_image(
 326         img.float() / 255.0, "/tmp/world.png", nrow=6, padding=6, pad_value=0
 327     )