# Dependencies, see also requirement.txt ;) import gradio as gr import cv2 import numpy as np import os from scenedetect import open_video, SceneManager from scenedetect.detectors import ContentDetector from moviepy.video.io.ffmpeg_tools import ffmpeg_extract_subclip def convert_to_tuple(list): return tuple(list); def clear_app(): return None, 27, None, None, None def find_scenes(video_path, threshold): # file name without extension filename = os.path.splitext(os.path.basename(video_path))[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() # Push the list of scenes into data_outputs data_outputs.append(scene_list) gradio_components_outputs.append("json") #print(scene_list) timecodes = [] if not scene_list: gr.Warning("No scenes detected in this video") return None, None, None timecodes.append({"title": filename + ".mp4", "fps": scene_list[0][0].get_framerate()}) shots = [] stills = [] # 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() / framerate tc_in = shot[0].get_timecode() tc_out = shot[1].get_timecode() frame_in = shot[0].get_frames() frame_out = shot[1].get_frames() timecode = {"tc_in": tc_in, "tc_out": tc_out, "frame_in": frame_in, "frame_out": frame_out} timecodes.append(timecode) # Set name template for each shot target_name = "shot_" + str(i+1) + "_" + str(filename) + ".mp4" # Split chunk ffmpeg_extract_subclip(video_path, shot_in, shot_out, targetname=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 vid = cv2.VideoCapture(video_path) fps = vid.get(cv2.CAP_PROP_FPS) print('frames per second =',fps) frame_id = shot[0].get_frames() # value from scene_list from step 1 vid.set(cv2.CAP_PROP_POS_FRAMES, frame_id) ret, frame = vid.read() # Save frame as PNG file img = str(frame_id) + '_screenshot.png' cv2.imwrite(img,frame) # Push image into stills List stills.append((img, 'shot ' + str(i+1))) # 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 timecodes, shots, stills # ————————————————————————————————————————————————— # 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 = [] #SET OUTPUTS # This would be nice if number of outputs could be set after Interface Launch: # because we do not know how many shots will be detected # gradio_components_outputs = [ "json", "video", "video", ... , "video", "file", "gallery" ] # outputs = gradio_components_outputs # ANOTHER SOLUTION WOULD BE USING A (FUTURE ?) "VIDEO GALLERY" GRADIO COMPONENT FROM LIST :) with gr.Blocks() as demo: with gr.Column(): gr.Markdown(""" # Scene Edit Detection Copy of @fffiloni's gradio demo of PySceneDetect. Automatically find all the shots in a video. Accepts mp4 format. Works only with videos that have cuts in them. """) with gr.Row(): with gr.Column(): video_input = gr.Video(sources="upload", format="mp4", label="Video Sequence", mirror_webcam = False) threshold = gr.Slider(label="Threshold pixel comparison: if exceeded, triggers a scene cut. Default: 27.0", minimum=15.0, maximum=40.0, value=27.0) with gr.Row(): clear_button = gr.Button(value=("Clear")) run_button = gr.Button(value = "Submit", variant = "primary") with gr.Column(): json_output = gr.JSON(label="Shots detected") file_output = gr.File(label="Downloadable Shots") gallery_output = gr.Gallery(label="Still Images from each shot", object_fit = "cover", columns = 3) run_button.click(fn=find_scenes, inputs=[video_input, threshold], outputs=[json_output, file_output, gallery_output]) clear_button.click(fn=clear_app, inputs = None, outputs=[video_input, threshold, json_output, file_output, gallery_output]) demo.queue().launch(debug=True)