adding fps options, and testing inference-1

gradio_app.py

@@ -206,7 +206,7 @@ download_models()
 # -----------------------------
 # Step 2: Inference Logic
 # -----------------------------
-def run_epic_inference(video_path, num_frames, target_pose, mode):
+def run_epic_inference(video_path, fps, num_frames, target_pose, mode):
     temp_input_path = "/app/temp_input.mp4"
     output_dir = "/app/output_anchor"
     video_output_path = f"{output_dir}/masked_videos/output.mp4"
@@ -218,7 +218,7 @@ def run_epic_inference(video_path, num_frames, target_pose, mode):
     try:
         theta, phi, r, x, y = target_pose.strip().split()
     except ValueError:
-        return f"❌ Invalid target pose format. Use: θ φ r x y", None
+        return f"Invalid target pose format. Use: θ φ r x y", None
     logs =  f"Running inference with target pose: θ={theta}, φ={phi}, r={r}, x={x}, y={y}\n"
     command = [
         "python", "/app/inference/v2v_data/inference.py",
@@ -232,8 +232,8 @@ def run_epic_inference(video_path, num_frames, target_pose, mode):
         "--video_length", str(num_frames),
         "--save_name", "output",
         "--mode", mode,
+        "--fps", fps
     ]
-
     try:
         result = subprocess.run(command, capture_output=True, text=True, check=True)
         logs += result.stdout
@@ -255,8 +255,8 @@ def print_output_directory(out_dir):
     return result
 
 def inference(video_path, num_frames, fps, target_pose, mode):
-    logs, video_masked = run_epic_inference(video_path, num_frames, target_pose, mode)
-
+    logs, video_masked = run_epic_inference(video_path, fps, num_frames, target_pose, mode)
+    return logs, video_masked
     result_dir = print_output_directory("/app/output_anchor")
 
 
@@ -291,7 +291,8 @@ def inference(video_path, num_frames, fps, target_pose, mode):
         "--controlnet_transformer_num_layers", "8",
         "--infer_with_mask",
         "--pool_style", "max",
-        "--seed", "43"
+        "--seed", "43",
+        "--fps", fps
     ]
 
     result = subprocess.run(command, capture_output=True, text=True)

inference/v2v_data/demo.py

@@ -180,8 +180,8 @@ class GetAnchorVideos:
         mask_save = process_mask_tensor(torch.cat(masks)).squeeze().cpu().numpy()
         np.save(f"{opts.out_dir}/depth/{save_name}.npy",depths.cpu().numpy())
         np.savez_compressed(f"{opts.out_dir}/masks/{save_name}.npz",mask=mask_save)
-        save_video_as_mp4(ori_video_save,f"{opts.out_dir}/videos/{save_name}.mp4", fps=8)
-        save_video_as_mp4(cond_video_save,f"{opts.out_dir}/masked_videos/{save_name}.mp4", fps=8)
+        save_video_as_mp4(ori_video_save,f"{opts.out_dir}/videos/{save_name}.mp4", fps=opts.fps)
+        save_video_as_mp4(cond_video_save,f"{opts.out_dir}/masked_videos/{save_name}.mp4", fps=opts.fps)
         np.save(f'{opts.out_dir}/post_t/' + save_name + '.npy',pose_t.cpu().numpy())
         np.save(f'{opts.out_dir}/pose_s/' + save_name + '.npy',pose_s.cpu().numpy())
         np.save(f'{opts.out_dir}/intrinsics/' + save_name + '.npy',K[0].cpu().numpy())