Update app.py

app.py

@@ -1,6 +1,5 @@
-import os, cv2, glob, time, torch, shutil
-import argparse, numpy as np, concurrent.futures
-import insightface, onnxruntime, gradio as gr
+import os, cv2, glob, time, torch, shutil, argparse, concurrent.futures
+import numpy as np, insightface, onnxruntime, gradio as gr
 from moviepy.editor import VideoFileClip
 from face_swapper import Inswapper, paste_to_whole
 from face_analyser import detect_conditions, get_analysed_data, swap_options_list
@@ -8,7 +7,7 @@ from face_parsing import init_parsing_model, get_parsed_mask, mask_regions, mask
 from face_enhancer import get_available_enhancer_names, load_face_enhancer_model, cv2_interpolations
 from utils import merge_img_sequence_from_ref, open_directory, split_list_by_lengths, create_image_grid
 
-# ----------------- USER ARGS -----------------
+# ---------------- ARGS ----------------
 parser = argparse.ArgumentParser(description="Free Face Swapper (2 faces)")
 parser.add_argument("--out_dir", default=os.getcwd())
 parser.add_argument("--batch_size", default=32)
@@ -21,7 +20,7 @@ DEF_OUTPUT_PATH, BATCH_SIZE = args.out_dir, int(args.batch_size)
 device = "cuda" if USE_CUDA else "cpu"
 EMPTY_CACHE = lambda: torch.cuda.empty_cache() if device == "cuda" else None
 
-# ----------------- MODELS -----------------
+# ---------------- MODELS ----------------
 PROVIDER = ["CUDAExecutionProvider", "CPUExecutionProvider"] if USE_CUDA else ["CPUExecutionProvider"]
 FACE_ANALYSER, FACE_SWAPPER, FACE_PARSER = None, None, None
 FACE_ENHANCER_LIST = ["NONE"] + get_available_enhancer_names() + cv2_interpolations
@@ -43,12 +42,11 @@ def load_face_parser_model(path="./assets/pretrained_models/79999_iter.pth"):
     if FACE_PARSER is None:
         FACE_PARSER = init_parsing_model(path, device=device)
 
-# ----------------- SWAP LOGIC -----------------
+# ---------------- SWAP ----------------
 def run_face_swap(image_sequence, source_path, label, condition, age, distance, face_scale,
                   face_enhancer_name, enable_face_parser, mask_includes,
                   mask_soft_kernel, mask_soft_iterations, blur_amount, erode_amount,
                   enable_laplacian_blend, crop_mask):
-    global PREVIEW
     yield f"### ⌛ {label}: phân tích khuôn mặt..."
     source_data = (source_path, age) if condition != "Specific Face" else (([], []), distance)
 
@@ -57,18 +55,18 @@ def run_face_swap(image_sequence, source_path, label, condition, age, distance,
         swap_condition=condition, detect_condition="best detection", scale=face_scale
     )
 
-    # Swap
+    # swap
     yield f"### ⌛ {label}: hoán đổi..."
     preds, matrs = [], []
     for bp, bm in FACE_SWAPPER.batch_forward(whole_frame_list, analysed_targets, analysed_sources):
         preds.extend(bp); matrs.extend(bm); EMPTY_CACHE()
 
-    # Enhance
+    # enhance
     if face_enhancer_name != "NONE":
         enhancer_model, enhancer_runner = load_face_enhancer_model(name=face_enhancer_name, device=device)
         preds = [cv2.resize(enhancer_runner(p, enhancer_model), (512, 512)) for p in preds]
 
-    # Mask
+    # mask
     masks = [None] * len(preds)
     if enable_face_parser:
         yield f"### ⌛ {label}: tạo mask..."
@@ -78,7 +76,7 @@ def run_face_swap(image_sequence, source_path, label, condition, age, distance,
             masks.append(bm); EMPTY_CACHE()
         masks = np.concatenate(masks, axis=0) if masks else [None] * len(preds)
 
-    # Paste back
+    # paste back
     yield f"### ⌛ {label}: dán khuôn mặt..."
     sp_preds = split_list_by_lengths(preds, num_faces_per_frame)
     sp_matrs = split_list_by_lengths(matrs, num_faces_per_frame)
@@ -98,7 +96,7 @@ def run_face_swap(image_sequence, source_path, label, condition, age, distance,
     with concurrent.futures.ThreadPoolExecutor() as exe:
         list(exe.map(post_process, range(len(image_sequence)), image_sequence))
 
-# ----------------- PROCESS MAIN -----------------
+# ---------------- PROCESS ----------------
 def process(input_type, image_path, video_path, directory_path,
             source_female_path, source_male_path,
             output_path, output_name, keep_output_sequence,
@@ -119,7 +117,7 @@ def process(input_type, image_path, video_path, directory_path,
         load_face_enhancer_model(name=face_enhancer_name, device=device)
     if enable_face_parser: load_face_parser_model()
 
-    # --- chuẩn bị input ---
+    # chuẩn bị input
     image_sequence, output_file = [], None
     if input_type == "Image":
         output_file = os.path.join(output_path, output_name + ".png")
@@ -145,7 +143,7 @@ def process(input_type, image_path, video_path, directory_path,
     else:
         yield "### ❌ Stream chưa hỗ trợ"; return
 
-    # --- Swap nữ, nam ---
+    # swap nữ / nam
     has_swap = False
     if source_female_path and os.path.exists(source_female_path):
         has_swap = True
@@ -181,50 +179,66 @@ def process(input_type, image_path, video_path, directory_path,
           gr.update(value=PREVIEW, visible=True), gr.update(interactive=True), \
           gr.update(interactive=True), gr.update(value=OUTPUT_FILE, visible=True)
 
-# ----------------- GUI -----------------
-with gr.Blocks(css="footer{display:none !important}") as interface:
+# ---------------- GUI ----------------
+css = """footer{display:none !important}"""
+
+with gr.Blocks(css=css) as interface:
     gr.Markdown("# 🗿 Free Face Swapper (2 faces)")
+
     with gr.Row():
-        with gr.Column(scale=0.4):
-            swap_option = gr.Dropdown(swap_options_list, value=swap_options_list[0], label="Swap Condition")
-            age = gr.Number(value=25, label="Age", visible=False)
-            output_directory = gr.Text(value=DEF_OUTPUT_PATH, label="Output Dir")
-            output_name = gr.Text(value="Result", label="Output Name")
-            keep_output_sequence = gr.Checkbox(label="Keep sequence", value=False)
-            face_scale = gr.Slider(0, 2, 1, label="Face Scale")
-            face_enhancer_name = gr.Dropdown(FACE_ENHANCER_LIST, value="NONE", label="Face Enhancer")
-            enable_face_parser_mask = gr.Checkbox(label="Enable Face Parsing", value=False)
-            mask_include = gr.Dropdown(mask_regions.keys(), value=list(mask_regions.keys()), multiselect=True, label="Mask Include")
-            mask_soft_kernel = gr.Number(value=17, label="Soft Kernel", visible=False)
-            mask_soft_iterations = gr.Number(value=10, label="Soft Iterations")
-            crop_top = gr.Slider(0, 511, 0, label="Crop Top")
-            crop_bott = gr.Slider(0, 511, 511, label="Crop Bottom")
-            crop_left = gr.Slider(0, 511, 0, label="Crop Left")
-            crop_right = gr.Slider(0, 511, 511, label="Crop Right")
-            erode_amount = gr.Slider(0, 1, 0.15, step=0.05, label="Mask Erode")
-            blur_amount = gr.Slider(0, 1, 0.1, step=0.05, label="Mask Blur")
-            enable_laplacian_blend = gr.Checkbox(label="Laplacian Blend", value=True)
-
-            gr.Markdown("### 👩 Source Female")
-            source_female_input = gr.Image(label="Source Female", type="filepath")
-            gr.Markdown("### 👨 Source Male")
-            source_male_input = gr.Image(label="Source Male", type="filepath")
-
-            input_type = gr.Radio(["Image", "Video", "Directory"], value="Image", label="Target Type")
-            image_input = gr.Image(label="Target Image", type="filepath", visible=True)
-            video_input = gr.Video(label="Target Video", visible=False)
-            direc_input = gr.Text(label="Target Directory", visible=False)
-
-        with gr.Column(scale=0.6):
-            info = gr.Markdown("...")
-            with gr.Row():
-                swap_button = gr.Button("✨ Swap", variant="primary")
-                cancel_button = gr.Button("⛔ Cancel")
-            preview_image = gr.Image(label="Output Image", interactive=False)
-            preview_video = gr.Video(label="Output Video", interactive=False, visible=False)
-            with gr.Row():
-                output_directory_button = gr.Button("📂", visible=False)
-                output_video_button = gr.Button("🎬", visible=False)
+        with gr.Row():
+            with gr.Column(scale=0.4):
+                with gr.Tab("📄 Swap Condition"):
+                    swap_option = gr.Dropdown(swap_options_list, value=swap_options_list[0], label="Condition")
+                    age = gr.Number(value=25, label="Age", visible=False)
+
+                with gr.Tab("🎚️ Detection Settings"):
+                    detect_condition_dropdown = gr.Dropdown(detect_conditions, value="best detection", label="Condition")
+                    detection_size = gr.Number(value=640, label="Detection Size")
+                    detection_threshold = gr.Number(value=0.6, label="Detection Threshold")
+                    apply_detection_settings = gr.Button("Apply settings")
+
+                with gr.Tab("📤 Output Settings"):
+                    output_directory = gr.Text(value=DEF_OUTPUT_PATH, label="Output Directory")
+                    output_name = gr.Text(value="Result", label="Output Name")
+                    keep_output_sequence = gr.Checkbox(label="Keep sequence", value=False)
+
+                with gr.Tab("🪄 Other Settings"):
+                    face_scale = gr.Slider(0, 2, 1, label="Face Scale")
+                    face_enhancer_name = gr.Dropdown(FACE_ENHANCER_LIST, value="NONE", label="Face Enhancer")
+                    enable_face_parser_mask = gr.Checkbox(label="Enable Face Parsing", value=False)
+                    mask_include = gr.Dropdown(mask_regions.keys(), value=list(mask_regions.keys()), multiselect=True)
+                    mask_soft_kernel = gr.Number(value=17, visible=False)
+                    mask_soft_iterations = gr.Number(value=10, label="Soft Iterations")
+                    crop_top = gr.Slider(0, 511, 0, label="Crop Top")
+                    crop_bott = gr.Slider(0, 511, 511, label="Crop Bottom")
+                    crop_left = gr.Slider(0, 511, 0, label="Crop Left")
+                    crop_right = gr.Slider(0, 511, 511, label="Crop Right")
+                    erode_amount = gr.Slider(0, 1, 0.15, step=0.05, label="Mask Erode")
+                    blur_amount = gr.Slider(0, 1, 0.1, step=0.05, label="Mask Blur")
+                    enable_laplacian_blend = gr.Checkbox(value=True, label="Laplacian Blend")
+
+                gr.Markdown("### 👩 Source Female")
+                source_female_input = gr.Image(label="Source Female Face", type="filepath")
+
+                gr.Markdown("### 👨 Source Male")
+                source_male_input = gr.Image(label="Source Male Face", type="filepath")
+
+                input_type = gr.Radio(["Image", "Video", "Directory"], value="Image", label="Target Type")
+                image_input = gr.Image(label="Target Image", type="filepath", visible=True)
+                video_input = gr.Video(label="Target Video", visible=False)
+                direc_input = gr.Text(label="Directory Path", visible=False)
+
+            with gr.Column(scale=0.6):
+                info = gr.Markdown("...")
+                with gr.Row():
+                    swap_button = gr.Button("✨ Swap", variant="primary")
+                    cancel_button = gr.Button("⛔ Cancel")
+                preview_image = gr.Image(label="Output", interactive=False)
+                preview_video = gr.Video(label="Output", interactive=False, visible=False)
+                with gr.Row():
+                    output_directory_button = gr.Button("📂", visible=False)
+                    output_video_button = gr.Button("🎬", visible=False)
 
     swap_inputs = [
         input_type, image_input, video_input, direc_input,