fix: refactor bug

src/ddpm/sample.py

@@ -1,9 +1,11 @@
+import argparse
 import torch
 import matplotlib
 import matplotlib.pyplot as plt
 import numpy as np
 from train import get_diffusion_params
-from train import TIMESTEPS, IMAGE_SIZE
+from diffusion import TIMESTEPS
+from values import EXPERIMENTS
 from model import UNet
 
 from generate_circle_dataset import (
@@ -12,12 +14,13 @@ from generate_circle_dataset import (
 
 from utils import compute_statistics, generate_synthetic_samples
 
+IMAGE_SIZE = 32
 matplotlib.use("Agg")
 
 torch.manual_seed(1)
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
-GENERATE_SYNTHETIC_DATA = True
+GENERATE_SYNTHETIC_DATA = False
 
 
 def load_dataset(data_path: str):
@@ -27,19 +30,19 @@ def load_dataset(data_path: str):
     return data
 
 
-if __name__ == "__main__":
-    torch.set_float32_matmul_precision("high")
-
+def main(args):
     plt.figure(figsize=(10, 10))
 
-    data = load_dataset("./data/circle_dataset/data-bigrectangle.npy")
+    data = load_dataset(f"./data/circle_dataset/data-{args.experiment_name}.npy")
 
     if GENERATE_SYNTHETIC_DATA:
         nb_synthetic_samples = 102_400
         params = get_diffusion_params(TIMESTEPS, device, eta=0.0)
 
         model = UNet(IMAGE_SIZE, TIMESTEPS).to(device)
-        model.load_state_dict(torch.load("model-bigrectangle.pkl", weights_only=True))
+        model.load_state_dict(
+            torch.load(f"model-{args.experiment_name}.pkl", weights_only=True)
+        )
 
         model.eval()
         generated = generate_synthetic_samples(
@@ -49,12 +52,34 @@ if __name__ == "__main__":
             params=params,
             device=device,
         )
-        np.save("./data/circle_dataset/generated-bigrectangle.npy", generated)
+        np.save(
+            f"./data/circle_dataset/generated-{args.experiment_name}.npy", generated
+        )
     else:
-        generated = np.load("./data/circle_dataset/generated-bigrectangle.npy")
+        generated = np.load(
+            f"./data/circle_dataset/generated-{args.experiment_name}.npy"
+        )
 
     visualize_samples(generated, "")
     compute_statistics(data, generated)
     # plot_k_centers(generated, 1)
     # plot_bad_centers(generated)
     # plot_more_than_2_centers(generated)
+
+
+if __name__ == "__main__":
+    torch.set_float32_matmul_precision("high")
+
+    parser = argparse.ArgumentParser(
+        desc="Generate samples from a trained diffusion model and save to disk."
+    )
+    parser.add_argument(
+        "-e",
+        "--experiment_name",
+        default="vanilla",
+        type=str,
+        choices=EXPERIMENTS.keys(),
+    )
+    args = parser.parse_args()
+
+    main(args)

src/ddpm/train.py

@@ -21,19 +21,28 @@ BATCH_SIZE = 512
 CHANNELS = 3
 
 # Histogram generation
-HIST_GENERATED_SAMPLES = 10_000
+HIST_GENERATED_SAMPLES = 10_112
 HIST_DATASET_SAMPLES = (
     100_000  # Using the whole dataset becomes super slow, so we take a subset
 )
 
 
-def load_dataset(data_path: str):
+def load_dataset(data_path: str, hist_set: int = 0):
     print("Loading dataset... ", end="", flush=True)
-    data = np.load(data_path).astype(np.float32)
-    data = data / 255.0  # normalize between [0-1]
+    data = np.load(data_path)
+
+    data_hist = None
+    if hist_set > 0:
+        hist_ids = np.random.choice(
+            np.arange(hist_set), replace=False, size=(hist_set,)
+        )
+        data_hist = data[hist_ids]
+
+    data = data.astype(np.float32) / 255.0  # normalize between [0-1]
     data = np.permute_dims(data, (0, 3, 1, 2))  # (b, h, w, c) -> (b, c, h, w)
     print("Done.")
-    return data
+
+    return data, data_hist
 
 
 def create_dataset_loader(data: np.array, batch_size: int):
@@ -111,11 +120,11 @@ def main(args):
     params = get_diffusion_params(TIMESTEPS, device)
     loss_fn = get_loss_fn(model, params)
 
-    data = load_dataset(f"./data/circle_dataset/data-{args.experiment_name}.npy")
-
-    if args.histogram:
-        ids = np.random.choice(np.arange(data.shape[0]), replace=False)
-        data_hist_set = data[ids]
+    data_hist_samples = HIST_DATASET_SAMPLES if args.histogram else None
+    data, data_hist = load_dataset(
+        f"./data/circle_dataset/data-{args.experiment_name}.npy",
+        data_hist_samples,
+    )
 
     # Main training loop
     train_losses = []
@@ -133,7 +142,7 @@ def main(args):
                 device=device,
             )
             compute_statistics(
-                data_hist_set,
+                data_hist,
                 generated,
                 output_path=f"center_statistics_{args.experiment_name}_{e}.png",
             )

src/ddpm/utils.py

@@ -39,6 +39,11 @@ def detect_circle_centers(image, circle_colors, threshold=30, min_pixels=12):
         centers (list of tuples): List of (row, col) coordinates for each detected circle.
         image (np.ndarray): The original image.
     """
+
+    image_orig = image
+    if image.shape[0] == 3:
+        image = np.permute_dims(image, (1, 2, 0))
+
     centers = []
     # Loop over each target circle color.
     for color in circle_colors:
@@ -61,7 +66,7 @@ def detect_circle_centers(image, circle_colors, threshold=30, min_pixels=12):
             center = center_of_mass(mask, labeled, i)
             centers.append(center)
 
-    return centers, image
+    return centers, image_orig
 
 
 def compute_statistics(
@@ -69,8 +74,6 @@ def compute_statistics(
     generated: np.array,
     output_path: str = "center_statistics.png",
 ):
-    assert len(data.shape) == 4
-    assert data.shape[2] == data.shape[3]
     image_size = data.shape[2]
 
     data_centers = []