feat(ml): debug inference_vid for canny

scripts/gen_vid_diffusion.py

@@ -275,6 +275,7 @@ def generate(
     bbox_select_list = []
     img_tensor_list = []
     out_img_list = []
+    sequence_count = 0
     for img_path, bbox_path in zip(limited_paths_img, limited_paths_bbox):
         img_in = os.path.join(os.path.dirname(os.path.dirname(paths_in_file)), img_path)
         bbox_in = os.path.join(
@@ -599,7 +600,7 @@ def generate(
             cond_image = fill_img_with_sketch(
                 img_tensor.unsqueeze(0), mask.unsqueeze(0)
             )
-        elif opt.alg_diffusion_cond_image_creation == "canny":
+        elif opt.alg_diffusion_cond_image_creation == "computed_sketch":  # "canny":
             clamp = torch.clamp(mask, 0, 1)
             if cond_in:
                 # mask the background to avoid canny edges around cond image
@@ -613,6 +614,7 @@ def generate(
                 high_threshold=alg_diffusion_sketch_canny_thresholds[1],
                 low_threshold_random=-1,
                 high_threshold_random=-1,
+                select_canny=[1] + [0] * (opt.data_temporal_number_frames - 1),
             )
             if cond_in:
                 # restore background
@@ -683,8 +685,22 @@ def generate(
         else:
             ref_tensor = None
 
-        cond_image_list.append(cond_image)
-        y_t_list.append(y_t)
+        if opt.alg_diffusion_cond_image_creation == "computed_sketch":
+            if sequence_count == 0:
+                cond_image_list.append(cond_image)
+                y_t_list.append(y_t)
+            else:
+                cond_image_list.append(y_t_list[0])
+                y_t_list.append(y_t_list[0])
+        if opt.alg_diffusion_cond_image_creation == "y_t":
+            if sequence_count == 0:
+                cond_image_list.append(cond_image)
+                y_t_list.append(y_t)
+            else:
+                cond_image_list.append(cond_image)
+                y_t_list.append(y_t_list[0])
+
+        sequence_count = sequence_count + 1
         y0_tensor_list.append(y0_tensor)
         mask_list.append(mask)
         bbox_select_list.append(bbox_select)