v0.9.2

readme.md

@@ -61,11 +61,12 @@ To install the extension go to 'Extensions' tab in [Automatic1111 web-ui](https:
 * The extension might work incorrectly if 'Apply color correction to img2img results to match original colors.' option is enabled. Make sure to disable it in 'Settings' tab -> 'Stable Diffusion' section. 
 
 ## Last version changes: v0.9
-* Fixed issues #69, #76, #91.
+* Fixed issues #69, #76, #91, #92.
 * Fixed an issue in vid2vid mode when an occlusion mask computed from the optical flow may include unnecessary parts (where flow is non-zero).
 * Added 'Extra params' in vid2vid mode for more fine-grain controls of the processing pipeline.
 * Better default parameters set for vid2vid pipeline.
 * In txt2vid mode after the first frame is generated the seed is now automatically set to -1 to prevent blurring issues.
 * Added an option to save resulting frames into a folder alongside the video.
 * Added ability to export current parameters in a human readable form as a json.
+* Interpolation mode in the flow-applying stage is set to ‘nearest’ to reduce overtime image blurring.
 

scripts/base_ui.py

@@ -196,7 +196,7 @@ def on_ui_tabs():
           
         with gr.Accordion("Export settings", open=False):
           export_settings_button = gr.Button('Export', elem_id=f"sdcn_export_settings_button")
-          export_setting_json = gr.JSON(value='{}')
+          export_setting_json = gr.Code(value='')
 
 
       with gr.Column(scale=1, variant='compact'):

scripts/core/flow_utils.py

@@ -116,8 +116,8 @@ def compute_diff_map(next_flow, prev_flow, prev_frame, cur_frame, prev_frame_sty
   prev_frame_torch = torch.from_numpy(prev_frame).float().unsqueeze(0).permute(0, 3, 1, 2) #N, C, H, W
   prev_frame_styled_torch = torch.from_numpy(prev_frame_styled).float().unsqueeze(0).permute(0, 3, 1, 2) #N, C, H, W
 
-  warped_frame = torch.nn.functional.grid_sample(prev_frame_torch, flow_grid, padding_mode="reflection").permute(0, 2, 3, 1)[0].numpy()
-  warped_frame_styled = torch.nn.functional.grid_sample(prev_frame_styled_torch, flow_grid, padding_mode="reflection").permute(0, 2, 3, 1)[0].numpy()
+  warped_frame = torch.nn.functional.grid_sample(prev_frame_torch, flow_grid, mode="nearest", padding_mode="reflection").permute(0, 2, 3, 1)[0].numpy()
+  warped_frame_styled = torch.nn.functional.grid_sample(prev_frame_styled_torch, flow_grid, mode="nearest", padding_mode="reflection").permute(0, 2, 3, 1)[0].numpy()
 
   #warped_frame = cv2.remap(prev_frame, flow_map, None, cv2.INTER_NEAREST, borderMode = cv2.BORDER_REFLECT)
   #warped_frame_styled = cv2.remap(prev_frame_styled, flow_map, None, cv2.INTER_NEAREST, borderMode = cv2.BORDER_REFLECT)

scripts/core/txt2vid.py

@@ -116,7 +116,7 @@ def start_process(*args):
       flow_map[:,:,0] += np.arange(args_dict['width'])
       flow_map[:,:,1] += np.arange(args_dict['height'])[:,np.newaxis]
 
-      warped_frame = cv2.remap(prev_frame, flow_map, None, cv2.INTER_CUBIC, borderMode = cv2.BORDER_REFLECT_101)
+      warped_frame = cv2.remap(prev_frame, flow_map, None, cv2.INTER_NEAREST, borderMode = cv2.BORDER_REFLECT_101)
 
       curr_frame = warped_frame.copy()
       

scripts/core/vid2vid.py

@@ -171,8 +171,8 @@ def start_process(*args):
         sdcn_anim_tmp.frames_prepared = False
 
         cn = sdcn_anim_tmp.process_counter % 10 
-        curr_frame = sdcn_anim_tmp.prepared_frames[cn+1]
-        prev_frame = sdcn_anim_tmp.prepared_frames[cn]
+        curr_frame = sdcn_anim_tmp.prepared_frames[cn+1][...,:3]
+        prev_frame = sdcn_anim_tmp.prepared_frames[cn][...,:3]
         next_flow = sdcn_anim_tmp.prepared_next_flows[cn]
         prev_flow = sdcn_anim_tmp.prepared_prev_flows[cn]
 
@@ -193,7 +193,7 @@ def start_process(*args):
         occlusion_mask = np.clip(alpha_mask * 255, 0, 255).astype(np.uint8)
 
         # fix warped styled frame from duplicated that occures on the places where flow is zero, but only because there is no place to get the color from
-        warped_styled_frame = curr_frame[...,:3].astype(float) * alpha_mask + warped_styled_frame[...,:3].astype(float) * (1 - alpha_mask)
+        warped_styled_frame = curr_frame.astype(float) * alpha_mask + warped_styled_frame.astype(float) * (1 - alpha_mask)
 
         # process current frame
         # TODO: convert args_dict into separate dict that stores only params necessery for img2img processing