import numpy as np import cv2 import os import insightface from insightface.app import FaceAnalysis from insightface.data import get_image as ins_get_image import gradio as gr theme = gr.themes.Default( font=['Helvetica', 'ui-sans-serif', 'system-ui', 'sans-serif'], font_mono=['IBM Plex Mono', 'ui-monospace', 'Consolas', 'monospace'], ).set( border_color_primary='#c5c5d2', button_large_padding='6px 12px', body_text_color_subdued='#484848', background_fill_secondary='#eaeaea' ) def load_face_analyser_model(name="buffalo_l"): global FACE_ANALYSER if FACE_ANALYSER is None: FACE_ANALYSER = insightface.app.FaceAnalysis(name=name, providers=PROVIDER) FACE_ANALYSER.prepare( ctx_id=0, det_size=(DETECT_SIZE, DETECT_SIZE), det_thresh=DETECT_THRESH ) def load_face_swapper_model(path="./assets/pretrained_models/inswapper_128.onnx"): global FACE_SWAPPER if FACE_SWAPPER is None: batch = int(BATCH_SIZE) if device == "cuda" else 1 FACE_SWAPPER = Inswapper(model_file=path, batch_size=batch, providers=PROVIDER) def load_face_parser_model(path="./assets/pretrained_models/79999_iter.pth"): global FACE_PARSER if FACE_PARSER is None: FACE_PARSER = init_parsing_model(path, device=device) load_face_analyser_model() load_face_swapper_model() def add_bbox_padding(bbox, margin=5): return [ bbox[0] - margin, bbox[1] - margin, bbox[2] + margin, bbox[3] + margin] def select_handler(img, evt: gr.SelectData): faces = app.get(img) faces = sorted(faces, key = lambda x : x.bbox[0]) cropped_image = [] face_index = -1 sel_face_index = 0 print("Coords: ", evt.index[0],evt.index[1]) for face in faces: box = face.bbox.astype(np.int32) face_index = face_index + 1 if point_in_box((box[0], box[1]),(box[2],box[3]),(evt.index[0],evt.index[1])) == True: # print("True ", face_index) # print("Bbox org: ", box) # Add ~25% margin to the box so the face is recognized correctly margin = int((box[2]-box[0]) * 0.35) box = add_bbox_padding(box,margin) box = np.clip(box,0,None) print("Bbox exp: ", box) sel_face_index = face_index cropped_image = img[box[1]:box[3],box[0]:box[2]] return cropped_image, sel_face_index def point_in_box(bl, tr, p) : if (p[0] > bl[0] and p[0] < tr[0] and p[1] > bl[1] and p[1] < tr[1]) : return True else: return False def get_faces(img): faces = app.get(img) faces = sorted(faces, key = lambda x : x.bbox[0]) #boxed_faces = app.draw_on(img, faces) #for i in range(len(faces)): # face = faces[i] # box = face.bbox.astype(np.int32) # cv2.putText(boxed_faces,'Face#:%d'%(i), (box[0]-1, box[3]+14),cv2.FONT_HERSHEY_COMPLEX,0.7,(0,0,255),2) return img, len(faces) def swap_face_fct(img_source,face_index,img_swap_face): faces = app.get(img_source) faces = sorted(faces, key = lambda x : x.bbox[0]) src_face = app.get(img_swap_face) src_face = sorted(src_face, key = lambda x : x.bbox[0]) #print("index:",faces) res = swapper.get(img_source, faces[face_index], src_face[0], paste_back=True) return res def swap_video_fct(video_path, output_path, source_face, destination_face, tolerance, preview=-1, progress=gr.Progress()): # Get the Destination Face parameters (the face which should be swapped) dest_face = app.get(destination_face) dest_face = sorted(dest_face, key = lambda x : x.bbox[0]) if(len(dest_face) == 0): print("💡 No dest face found") return -1 dest_face_feats = [] dest_face_feats.append(dest_face[0].normed_embedding) dest_face_feats = np.array(dest_face_feats, dtype=np.float32) # Get the source face parameters (the face that replaces the original) src_face = app.get(source_face) src_face = sorted(src_face, key = lambda x : x.bbox[0]) if(len(src_face) == 0): print("🚨 No source face found") return -1 cap = cv2.VideoCapture(video_path) ret, frame = cap.read() fps = int(cap.get(cv2.CAP_PROP_FPS)) frame_count = int(cap.get(cv2.CAP_PROP_FRAME_COUNT)) width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) fourcc = cv2.VideoWriter_fourcc(*'avc1') # Use the same tmp dir from gradio if no output path is set if(len(output_path) > 0): out_path = output_path else: out_path = os.path.dirname(video_path) + "/out.mp4" if preview == -1: for_range = range(frame_count) video_out = cv2.VideoWriter(out_path,fourcc,fps,(width,height)) else: for_range = range(preview-1,preview) for i in for_range: progress(i/frame_count, desc="⏳Processing") cap.set(cv2.CAP_PROP_POS_FRAMES, i) ret, frame = cap.read() #frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) # Find all faces in the current frame faces = app.get(frame) faces = sorted(faces, key = lambda x : x.bbox[0]) # No face in Scene => copy input frame if(len(faces) > 0): feats = [] for face in faces: feats.append(face.normed_embedding) feats = np.array(feats, dtype=np.float32) sims = np.dot(dest_face_feats, feats.T) print(sims) # find the index of the most similar face max_index = np.argmax(sims) print("Sim:", max_index) if(sims[0][max_index]*100 >= (100-tolerance)): frame = swapper.get(frame, faces[max_index], src_face[0], paste_back=True) if preview == -1: video_out.write(frame) if preview == -1: video_out.release() return out_path else: return cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) ins_get_image def analyze_video(video_path): cap = cv2.VideoCapture(video_path) fps = int(cap.get(cv2.CAP_PROP_FPS)) frame_count = int(cap.get(cv2.CAP_PROP_FRAME_COUNT)) length = frame_count/fps width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) return f"Resolution: {width}x{height}\nLength: {length}\nFps: {fps}\nFrames: {frame_count}" def update_slider(video_path): cap = cv2.VideoCapture(video_path) fps = int(cap.get(cv2.CAP_PROP_FPS)) frame_count = int(cap.get(cv2.CAP_PROP_FRAME_COUNT)) length = frame_count/fps return gr.update(minimum=0,maximum=frame_count,value=frame_count/2) def show_preview(video_path, frame_number): cap = cv2.VideoCapture(video_path) cap.set(cv2.CAP_PROP_POS_FRAMES, frame_number) ret, frame = cap.read() frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) return frame def create_interface(): title = '# 🧸FaceSwap UI' with gr.Blocks(theme='WeixuanYuan/Base_dark') as face_swap_ui: with gr.Tab("📀 Swap Face Image"): with gr.Row(): with gr.Column(): image_input = gr.Image(label='🎞️Input Image (📺 Click to select a face)', scale=0.5) with gr.Row(): analyze_button = gr.Button("⌛ Analyze") with gr.Row(): with gr.Column(): face_num = gr.Number(label='📹 Recognized Faces') face_index_num = gr.Number(label='📀 Face Index', precision=0) selected_face = gr.Image(label='💊Face to swap', interactive=False) swap_face = gr.Image(label='🧪Swap Face') swap_button = gr.Button("🧩 Swap") with gr.Column(): image_output = gr.Image(label='📤Output Image',interactive=False) #text_output = gr.Textbox(placeholder="What is your name?") swap_button.click(fn=swap_face_fct, inputs=[image_input, face_index_num, swap_face], outputs=[image_output]) image_input.select(select_handler, image_input, [selected_face, face_index_num]) analyze_button.click(fn=get_faces, inputs=image_input, outputs=[image_input,face_num]) with gr.Tab("📺Swap Face Video"): with gr.Row(): with gr.Column(): source_video = gr.Video() video_info = gr.Textbox(label="📡Video Information") gr.Markdown("🛠️Select a frame for preview with the slider. Then select the face which should be swapped by clicking on it with the cursor") video_position = gr.Slider(label="📐Frame preview",interactive=True) frame_preview = gr.Image(label="📏Frame preview") face_index = gr.Textbox(label="📉Face-Index",interactive=False) with gr.Row(): dest_face_vid = gr.Image(label="👑Face to swap",interactive=True) source_face_vid = gr.Image(label="🔮New Face") gr.Markdown("🔑The higher the tolerance the more likely a wrong face will be swapped. 30-40 is a good starting point.") face_tolerance = gr.Slider(label="⏳Tolerance",value=40,interactive=True) preview_video = gr.Button("🪞Preview") video_file_path = gr.Text(label="🗳️Output Video path incl. file.mp4 (when left empty it will be put in the gradio temp dir)") process_video = gr.Button("⌛Process") with gr.Column(): with gr.Column(scale=1): image_output = gr.Image() output_video = gr.Video(interactive=False) with gr.Column(scale=1): pass # Component Events source_video.upload(fn=analyze_video,inputs=source_video,outputs=video_info) video_info.change(fn=update_slider,inputs=source_video,outputs=video_position) #preview_button.click(fn=show_preview,inputs=[source_video, video_position],outputs=frame_preview) frame_preview.select(select_handler, frame_preview, [dest_face_vid, face_index ]) video_position.change(show_preview,inputs=[source_video, video_position],outputs=frame_preview) process_video.click(fn=swap_video_fct,inputs=[source_video,video_file_path,source_face_vid,dest_face_vid, face_tolerance], outputs=output_video) preview_video.click(fn=swap_video_fct,inputs=[source_video,video_file_path,source_face_vid,dest_face_vid, face_tolerance, video_position], outputs=image_output) set_slider_range_event = set_slider_range_btn.click( video_changed, inputs=[video_input], outputs=[start_frame, end_frame, video_fps], ) trim_and_reload_event = trim_and_reload_btn.click( fn=trim_and_reload, inputs=[video_input, output_directory, output_name, start_frame, end_frame], outputs=[video_input, info], ) start_frame_event = start_frame.release( fn=slider_changed, inputs=[show_trim_preview_btn, video_input, start_frame], outputs=[preview_image, preview_video], show_progress=True, ) end_frame_event = end_frame.release( fn=slider_changed, inputs=[show_trim_preview_btn, video_input, end_frame], outputs=[preview_image, preview_video], show_progress=True, ) input_type.change( update_radio, inputs=[input_type], outputs=[input_image_group, input_video_group, input_directory_group], ) swap_option.change( swap_option_changed, inputs=[swap_option], outputs=[age, specific_face, source_image_input], ) apply_detection_settings.click( analyse_settings_changed, inputs=[detect_condition_dropdown, detection_size, detection_threshold], outputs=[info], ) src_specific_inputs = [] gen_variable_txt = ",".join( [f"src{i+1}" for i in range(NUM_OF_SRC_SPECIFIC)] + [f"trg{i+1}" for i in range(NUM_OF_SRC_SPECIFIC)] ) exec(f"src_specific_inputs = ({gen_variable_txt})") swap_inputs = [ input_type, image_input, video_input, direc_input, source_image_input, output_directory, output_name, keep_output_sequence, swap_option, age, distance_slider, face_enhancer_name, enable_face_parser_mask, mask_include, mask_soft_kernel, mask_soft_iterations, blur_amount, erode_amount, face_scale, enable_laplacian_blend, crop_top, crop_bott, crop_left, crop_right, *src_specific_inputs, ] swap_outputs = [ info, preview_image, output_directory_button, output_video_button, preview_video, ] swap_event = swap_button.click( fn=process, inputs=swap_inputs, outputs=swap_outputs, show_progress=True ) cancel_button.click( fn=stop_running, inputs=None, outputs=[info], cancels=[ swap_event, trim_and_reload_event, set_slider_range_event, start_frame_event, end_frame_event, ], show_progress=True, ) output_directory_button.click( lambda: open_directory(path=WORKSPACE), inputs=None, outputs=None ) output_video_button.click( lambda: open_directory(path=OUTPUT_FILE), inputs=None, outputs=None ) if __name__ == "__main__": if USE_COLAB: print("Running in colab mode") interface.queue(concurrency_count=2, max_size=20).launch(share=False) if __name__ == "__main__": app = FaceAnalysis(name='buffalo_l') app.prepare(ctx_id=0, det_size=(640, 640)) swapper = insightface.model_zoo.get_model('inswapper_128.onnx', download=True, download_zip=True) create_interface()