Update app.py

app.py

@@ -1,73 +1,126 @@
+# Dependencies, see also requirement.txt ;)
 import gradio as gr
 import cv2 
 import numpy as np
 
 from scenedetect import open_video, SceneManager
 from scenedetect.detectors import ContentDetector
-#from scenedetect.video_splitter import split_video_ffmpeg
 
 from moviepy.video.io.ffmpeg_tools import ffmpeg_extract_subclip
 
-outputs = ["json", "file", "gallery"]
+# —————————————————————————————————————————————————
+
+title = "Scene Edit Detection"
+description = "Gradio demo of PyScene scenedetect, to automatically find every shots in a video sequence, then save each shots as a splitted mp4 video chunk to download"
+
+# SET INPUTS
+video_input = gr.Video(source="upload", format="mp4");
+
+# SET DATA AND COMPONENTS OUTPUTS
+# This would be filled like this:
+# data_outputs = [ [List from detection], "video_chunk_n0.mp4", "video_chunk_n1.mp4", ... , "video_chunk_n.mp4", [List of video filepath to download], [List of still images from each shot found] ]
 data_outputs = []
 
+# This would be filled like this:
+# gradio_components_outputs = [ "json", "video", "video", ... , "video", "file", "gallery" ]
+gradio_components_outputs = []
 
-def fn(list):
-    return tuple(list);
+# This would be nice if number of outputs could be set after Interface Launch:         
+# gradio_components_outputs = [ "json", "video", "video", ... , "video", "file", "gallery" ]
+# outputs = gradio_components_outputs
+working_outputs = ["json", "file", "gallery"]
 
+# —————————————————————————————————————————————————
+
+def convert_to_tuple(list):
+    return tuple(list);
 
 
 def find_scenes(video_path, threshold=27.0):
+    
     # Open our video, create a scene manager, and add a detector.
     video = open_video(video_path)
     scene_manager = SceneManager()
     scene_manager.add_detector(
         ContentDetector(threshold=threshold))
+    
+    # Start detection 
     scene_manager.detect_scenes(video, show_progress=True)
     scene_list = scene_manager.get_scene_list()
     
-    #outputs.append("json")
+    # Push the list of scenes into data_outputs
     data_outputs.append(scene_list)
+    gradio_components_outputs.append("json")
     #print(scene_list)
     
     shots = []
     stills = []
     
-    for i, scene in enumerate(scene_list):
-        shot_in = scene[0].get_frames() / scene[0].get_framerate()
-        shot_out = (scene[1].get_frames() - 1) / scene[0].get_framerate()
+    # For each shot found, set entry and exit points as seconds from frame number
+    # Then split video into chunks and store them into shots List
+    # Then extract first frame of each shot as thumbnail for the gallery
+    for i, shot in enumerate(scene_list):
+        
+        # STEP 1
+        # Get timecode in seconds
+        framerate = shot[0].get_framerate()
+        shot_in = shot[0].get_frames() / framerate
+        shot_out = (shot[1].get_frames() - 1) / framerate
+        
+        # Set name template for each shot
         target_name = str(i)+"_cut.mp4"
+        
+        # Split chunk
         ffmpeg_extract_subclip(video_path, shot_in, shot_out, targetname=target_name)
-        data_outputs.append(target_name)
+        
+        # Push chunk into shots List
         shots.append(target_name)
         
+        # Push each chunk into data_outputs
+        data_outputs.append(target_name)
+        gradio_components_outputs.append("video")
+        
+        # —————————————————————————————————————————————————
+        
+        # STEP 2
+        # extract first frame of each shot with cv2
         video = cv2.VideoCapture(video_path)
-
         fps = video.get(cv2.CAP_PROP_FPS)
         print('frames per second =',fps)
-
-        frame_id = scene[0].get_frames()
+        
+        frame_id = shot[0].get_frames() # value from scene_list from step 1
 
         video.set(cv2.CAP_PROP_POS_FRAMES, frame_id)
         ret, frame = video.read()
 
-        # Display and save frame
-        #cv2.imshow('frame', frame); cv2.waitKey(0)
+        # Save frame as PNG file
         img = str(frame_id) + '_screenshot.png'
         cv2.imwrite(img,frame)
+        
+        # Push image into stills List
         stills.append(img)
-        #outputs.append("video")
-    #shot_in = scene_list[1][0].get_frames() / scene_list[1][0].get_framerate()
-    #shot_out = (scene_list[1][1].get_frames() - 1) / scene_list[1][0].get_framerate()
-    #print(shot_in, shot_out)
     
-    results = fn(data_outputs)
-    print(results)
-    print(stills)
-    #ffmpeg_extract_subclip(video_path, shot_in, shot_out, targetname="cut.mp4")
+    # Push the list of video shots into data_outputs for Gradio file component
+    data_outputs.append(shots)
+    gradio_components_outputs.append("file")
+        
+    # Push the list of still images into data_outputs
+    data_outputs.append(stills)
+    gradio_components_outputs.append("gallery")
     
+    # This would have been used as gradio outputs, 
+    # if we could set number of outputs after the interface launch
+    # That's not (yet ?) possible
+    results = convert_to_tuple(data_outputs)    
+    print(results)
+ 
+    # return List of shots as JSON, List of video chunks, List of still images  
+    # *
+    # Would be nice to be able to return my results tuple as outputs, 
+    # while number of chunks found is not fixed:
+    # return results 
     return scene_list, shots, stills
-    
-video_input=gr.Video(source="upload", format="mp4");
 
-gr.Interface(fn=find_scenes, inputs=video_input, outputs=outputs).launch()
+# —————————————————————————————————————————————————
+
+gr.Interface(fn=find_scenes, inputs=video_input, outputs=working_outputs, title=title, description=description).launch()