# Written by Francois Fleuret <francois@fleuret.org>
-import math, sys, tqdm, os
+import math, sys, tqdm, os, warnings
import torch, torchvision
token_background = 0
first_bird_token = 1
nb_bird_tokens = colors.size(0) - 1
- token_forward = first_bird_token + nb_bird_tokens
- token_backward = token_forward + 1
token2char = (
"_" + "".join([chr(ord("A") + n) for n in range(len(colors) - 1)]) + "><"
speed=2,
nb_iterations=2,
avoid_collision=True,
+ max_nb_cached_chunks=None,
+ chunk_size=None,
+ nb_threads=-1,
):
+ super().__init__(max_nb_cached_chunks, chunk_size, nb_threads)
self.height = height
self.width = width
self.nb_birds = nb_birds
self.nb_iterations = nb_iterations
self.avoid_collision = avoid_collision
- def direction_tokens(self):
- return self.token_forward, self.token_backward
-
def generate_frame_sequences(self, nb):
frame_sequences = []
######################################################################
- def generate_prompts_and_answers(self, nb):
- frame_sequences = self.generate_frame_sequences(nb)
- prompts = frame_sequences[:, : frame_sequences.size(0) // 2].flatten(1)
- answers = frame_sequences[:, frame_sequences.size(0) // 2 :].flatten(1)
- return prompts, answers
-
- def generate_token_sequences(self, nb):
- frame_sequences = self.generate_frame_sequences(nb)
-
- result = []
-
- for frame_sequence in frame_sequences:
- a = []
- if torch.rand(1) < 0.5:
- for frame in frame_sequence:
- if len(a) > 0:
- a.append(torch.tensor([self.token_forward]))
- a.append(frame.flatten())
- else:
- for frame in reversed(frame_sequence):
- if len(a) > 0:
- a.append(torch.tensor([self.token_backward]))
- a.append(frame.flatten())
-
- result.append(torch.cat(a, dim=0)[None, :])
-
- return torch.cat(result, dim=0)
-
- ######################################################################
-
def frame2img(self, x, scale=15):
- x = x.reshape(-1, self.height, self.width)
+ x = x.reshape(x.size(0), self.height, -1)
m = torch.logical_and(
x >= 0, x < self.first_bird_token + self.nb_bird_tokens
).long()
return x
- def seq2img(self, seq, scale=15):
- all = [
- self.frame2img(
- seq[:, : self.height * self.width].reshape(-1, self.height, self.width),
- scale,
- )
- ]
+ def seq2str(self, seq):
+ result = []
+ for s in seq:
+ result.append("".join([self.token2char[v] for v in s]))
+ return result
+
+ def save_image(
+ self,
+ result_dir,
+ filename,
+ prompts,
+ answers,
+ predicted_prompts=None,
+ predicted_answers=None,
+ ):
+ if predicted_prompts is None:
+ predicted_prompts = 255
- separator = torch.full((seq.size(0), 3, self.height * scale - 1, 1), 0)
+ if predicted_answers is None:
+ predicted_answers = 255
- t = self.height * self.width
+ def add_frame(x, c, margin, bottom=False):
+ if bottom:
+ h, w, di, dj = x.size(2) + margin, x.size(3), 0, 0
+ else:
+ h, w, di, dj = (
+ x.size(2) + 2 * margin,
+ x.size(3) + 2 * margin,
+ margin,
+ margin,
+ )
- while t < seq.size(1):
- direction_tokens = seq[:, t]
- t += 1
+ y = x.new_full((x.size(0), x.size(1), h, w), 0)
- direction_images = self.colors[
- torch.full(
- (direction_tokens.size(0), self.height * scale - 1, scale), 0
+ if type(c) is int:
+ y[...] = c
+ else:
+ c = c.long()[:, None]
+ c = (
+ (c == 1).long() * torch.tensor([0, 255, 0], device=c.device)
+ + (c == 0).long() * torch.tensor([255, 255, 255], device=c.device)
+ + (c == -1).long() * torch.tensor([255, 0, 0], device=c.device)
)
- ].permute(0, 3, 1, 2)
-
- for n in range(direction_tokens.size(0)):
- if direction_tokens[n] == self.token_forward:
- for k in range(scale):
- for l in [0, 1]:
- direction_images[
- n,
- :,
- (self.height * scale) // 2 - scale // 2 + k - l,
- 3 + scale // 2 - abs(k - scale // 2),
- ] = 0
- elif direction_tokens[n] == self.token_backward:
- for k in range(scale):
- for l in [0, 1]:
- direction_images[
- n,
- :,
- (self.height * scale) // 2 - scale // 2 + k - l,
- 3 + abs(k - scale // 2),
- ] = 0
- else:
- for k in range(2, scale - 2):
- for l in [0, 1]:
- direction_images[
- n,
- :,
- (self.height * scale) // 2 - scale // 2 + k - l,
- k,
- ] = 0
- direction_images[
- n,
- :,
- (self.height * scale) // 2 - scale // 2 + k - l,
- scale - 1 - k,
- ] = 0
-
- all += [
- separator,
- direction_images,
- separator,
- self.frame2img(
- seq[:, t : t + self.height * self.width].reshape(
- -1, self.height, self.width
- ),
- scale,
- ),
- ]
-
- t += self.height * self.width
-
- return torch.cat(all, dim=3)
+ y[...] = c[:, :, None, None]
- def seq2str(self, seq):
- result = []
- for s in seq:
- result.append("".join([self.token2char[v] for v in s]))
- return result
+ y[:, :, di : di + x.size(2), dj : dj + x.size(3)] = x
+
+ return y
+
+ margin = 4
+
+ img_prompts = add_frame(self.frame2img(prompts.to("cpu")), c=0, margin=1)
+ h = img_prompts.size(2)
+ img_answers = add_frame(self.frame2img(answers.to("cpu")), c=0, margin=1)
+
+ img_prompts = add_frame(img_prompts, c=255, margin=margin, bottom=True)
+ img_answers = add_frame(img_answers, c=255, margin=margin, bottom=True)
+
+ img_prompts = add_frame(
+ img_prompts, c=predicted_prompts, margin=margin, bottom=True
+ )
+ img_answers = add_frame(
+ img_answers, c=predicted_answers, margin=margin, bottom=True
+ )
+
+ marker_size = 16
+
+ separator = img_prompts.new_full(
+ (
+ img_prompts.size(0),
+ img_prompts.size(1),
+ img_prompts.size(2),
+ marker_size,
+ ),
+ 255,
+ )
+
+ separator[:, :, 0] = 0
+ separator[:, :, h - 1] = 0
+
+ for k in range(1, 2 * marker_size - 8):
+ i = k - (marker_size - 4)
+ j = marker_size - 5 - abs(i)
+ separator[:, :, h // 2 - 1 + i, 2 + j] = 0
+ separator[:, :, h // 2 - 1 + i + 1, 2 + j] = 0
+
+ img = torch.cat([img_prompts, separator, img_answers], dim=3)
- def save_image(self, input, result_dir, filename):
- img = self.seq2img(input.to("cpu"))
image_name = os.path.join(result_dir, filename)
- torchvision.utils.save_image(img.float() / 255.0, image_name, nrow=6, padding=4)
+ torchvision.utils.save_image(
+ img.float() / 255.0, image_name, nrow=6, padding=margin * 4, pad_value=1.0
+ )
+
+ ######################################################################
+
+ def nb_token_values(self):
+ return len(self.colors)
- def save_quizzes(self, input, result_dir, filename_prefix):
- self.save_image(input, result_dir, filename_prefix + ".png")
+ def generate_prompts_and_answers(self, nb):
+ frame_sequences = self.generate_frame_sequences(nb)
+ frame_sequences = torch.cat([x[None] for x in frame_sequences], dim=0)
+
+ prompts = frame_sequences[:, : frame_sequences.size(1) // 2].flatten(1)
+
+ answers = frame_sequences[:, frame_sequences.size(1) // 2 :].flatten(1)
+
+ # warnings.warn("dirty test with longer answer", RuntimeWarning)
+ # answers = torch.cat(
+ # [
+ # frame_sequences[:, frame_sequences.size(1) // 2 :],
+ # frame_sequences[:, frame_sequences.size(1) // 2 :],
+ # ],
+ # dim=3,
+ # ).flatten(1)
+
+ return prompts, answers
+
+ def save_quizzes(
+ self,
+ result_dir,
+ filename_prefix,
+ prompts,
+ answers,
+ predicted_prompts=None,
+ predicted_answers=None,
+ ):
+ self.save_image(
+ result_dir,
+ filename_prefix + ".png",
+ prompts,
+ answers,
+ predicted_prompts,
+ predicted_answers,
+ )
######################################################################
if __name__ == "__main__":
import time
- sky = Sky(height=6, width=8, speed=4, nb_iterations=2)
+ sky = Sky(height=6, width=8, speed=1, nb_iterations=4)
- start_time = time.perf_counter()
- token_sequences = sky.generate_token_sequences(nb=64)
- delay = time.perf_counter() - start_time
- print(f"{token_sequences.size(0)/delay:02f} seq/s")
+ prompts, answers = sky.generate_prompts_and_answers(4)
+
+ predicted_prompts = torch.randint(3, (prompts.size(0),)) - 1
+ predicted_answers = torch.randint(3, (prompts.size(0),)) - 1
+
+ sky.save_quizzes(
+ "/tmp", "test", prompts, answers, predicted_prompts, predicted_answers
+ )
+
+ # start_time = time.perf_counter()
+ # token_sequences = sky.generate_token_sequences(nb=64)
+ # delay = time.perf_counter() - start_time
+ # print(f"{token_sequences.size(0)/delay:02f} seq/s")
# print(sky.seq2str(seq[:4]))
# seq = (1 - m) * seq + m * 23
# print(seq.size())
- img = sky.seq2img(token_sequences)
+ # img = sky.seq2img(token_sequences)
# print(img.size())
- torchvision.utils.save_image(
- img.float() / 255.0, "/tmp/world.png", nrow=6, padding=6, pad_value=0
- )
+ # torchvision.utils.save_image(
+ # img.float() / 255.0, "/tmp/world.png", nrow=6, padding=6, pad_value=0
+ # )