release a lora version

README.md

@@ -4,11 +4,13 @@ An state-of-the-art virtual try-on solution that combines the power of [CATVTON]
 Also inspired by [In-Context LoRA](https://arxiv.org/abs/2410.23775) for prompt engineering.
 
 ## Update
-[![SOTA](https://img.shields.io/badge/SOTA-FID%205.59-brightgreen)](https://drive.google.com/file/d/1T2W5R1xH_uszGVD8p6UUAtWyx43rxGmI/view?usp=sharing)
-[![Dataset](https://img.shields.io/badge/Dataset-VITON--HD-blue)](https://github.com/shadow2496/VITON-HD)
 
 ---
-**Latest Achievement** (2024/11/24):
+**Latest Achievement** 
+(2024/11/25):
+- Released lora weights. 
+
+(2024/11/24):
 - Released FID score and gradio demo
 - CatVton-Flux-Alpha achieved **SOTA** performance with FID: `5.593255043029785` on VITON-HD dataset. Test configuration: scale 30, step 30. My VITON-HD test inferencing results available [here](https://drive.google.com/file/d/1T2W5R1xH_uszGVD8p6UUAtWyx43rxGmI/view?usp=sharing)
 
@@ -22,8 +24,8 @@ Also inspired by [In-Context LoRA](https://arxiv.org/abs/2410.23775) for prompt
 | ![Original](example/person/00008_00.jpg) | ![Garment](example/garment/00034_00.jpg) | ![Result](example/result/3.png) |
 
 ## Model Weights
-Hugging Face: 🤗 [catvton-flux-alpha](https://huggingface.co/xiaozaa/catvton-flux-alpha)
-
+LORA weights in Hugging Face: 🤗 [catvton-flux-alpha](https://huggingface.co/xiaozaa/catvton-flux-alpha)
+Fine-tuning weights in Hugging Face: 🤗 [catvton-flux-lora-alpha](https://huggingface.co/xiaozaa/catvton-flux-lora-alpha)
 The model weights are trained on the [VITON-HD](https://github.com/shadow2496/VITON-HD) dataset.
 
 ## Prerequisites
@@ -40,6 +42,19 @@ huggingface-cli login
 ## Usage
 
 Run the following command to try on an image:
+
+LORA version:
+```bash
+python tryon_inference_lora.py \
+--image ./example/person/00008_00.jpg \
+--mask ./example/person/00008_00_mask.png \
+--garment ./example/garment/00034_00.jpg \
+--seed 4096 \
+--output_tryon test_lora.png \
+--steps 30
+```
+
+Fine-tuning version:
 ```bash
 python tryon_inference.py \
 --image ./example/person/00008_00.jpg \
@@ -64,7 +79,8 @@ Gradio demo:
 - [x] Release the FID score
 - [x] Add gradio demo
 - [ ] Release updated weights with better performance
-- [ ] Train a smaller model
+- [x] Train a smaller model
+- [ ] Support comfyui
 
 ## Citation
 

tryon_inference_lora.py

@@ -0,0 +1,134 @@
+import argparse
+import torch
+from diffusers.utils import load_image, check_min_version
+from diffusers import FluxPriorReduxPipeline, FluxFillPipeline
+from diffusers import FluxTransformer2DModel
+import numpy as np
+from torchvision import transforms
+
+def run_inference(
+    image_path,
+    mask_path,
+    garment_path,
+    size=(576, 768),
+    num_steps=50,
+    guidance_scale=30,
+    seed=42,
+    pipe=None
+):
+    # Build pipeline
+    if pipe is None:
+        transformer = FluxTransformer2DModel.from_pretrained(
+            "xiaozaa/flux1-fill-dev-diffusers",   ## The official Flux-Fill weights
+            torch_dtype=torch.bfloat16
+        )
+        print("Start loading LoRA weights")
+        state_dict, network_alphas = FluxFillPipeline.lora_state_dict(
+            pretrained_model_name_or_path_or_dict="xiaozaa/catvton-flux-lora-alpha",     ## The tryon Lora weights
+            weight_name="pytorch_lora_weights.safetensors",
+            return_alphas=True
+        )
+        is_correct_format = all("lora" in key or "dora_scale" in key for key in state_dict.keys())
+        if not is_correct_format:
+            raise ValueError("Invalid LoRA checkpoint.")
+
+        FluxFillPipeline.load_lora_into_transformer(
+            state_dict=state_dict,
+            network_alphas=network_alphas,
+            transformer=transformer,
+        )
+        
+        pipe = FluxFillPipeline.from_pretrained(
+            "black-forest-labs/FLUX.1-dev",
+            transformer=transformer,
+            torch_dtype=torch.bfloat16
+        ).to("cuda")
+    else:
+        pipe.to("cuda")
+
+    pipe.transformer.to(torch.bfloat16)
+
+    # Add transform
+    transform = transforms.Compose([
+        transforms.ToTensor(),
+        transforms.Normalize([0.5], [0.5])  # For RGB images
+    ])
+    mask_transform = transforms.Compose([
+        transforms.ToTensor()
+    ])
+
+    # Load and process images
+    # print("image_path", image_path)
+    image = load_image(image_path).convert("RGB").resize(size)
+    mask = load_image(mask_path).convert("RGB").resize(size)
+    garment = load_image(garment_path).convert("RGB").resize(size)
+
+    # Transform images using the new preprocessing
+    image_tensor = transform(image)
+    mask_tensor = mask_transform(mask)[:1]  # Take only first channel
+    garment_tensor = transform(garment)
+
+    # Create concatenated images
+    inpaint_image = torch.cat([garment_tensor, image_tensor], dim=2)  # Concatenate along width
+    garment_mask = torch.zeros_like(mask_tensor)
+    extended_mask = torch.cat([garment_mask, mask_tensor], dim=2)
+
+    prompt = f"The pair of images highlights a clothing and its styling on a model, high resolution, 4K, 8K; " \
+            f"[IMAGE1] Detailed product shot of a clothing" \
+            f"[IMAGE2] The same cloth is worn by a model in a lifestyle setting."
+
+    generator = torch.Generator(device="cuda").manual_seed(seed)
+
+    result = pipe(
+        height=size[1],
+        width=size[0] * 2,
+        image=inpaint_image,
+        mask_image=extended_mask,
+        num_inference_steps=num_steps,
+        generator=generator,
+        max_sequence_length=512,
+        guidance_scale=guidance_scale,
+        prompt=prompt,
+    ).images[0]
+
+    # Split and save results
+    width = size[0]
+    garment_result = result.crop((0, 0, width, size[1]))
+    tryon_result = result.crop((width, 0, width * 2, size[1]))
+    
+    return garment_result, tryon_result
+
+def main():
+    parser = argparse.ArgumentParser(description='Run FLUX virtual try-on inference')
+    parser.add_argument('--image', required=True, help='Path to the model image')
+    parser.add_argument('--mask', required=True, help='Path to the agnostic mask')
+    parser.add_argument('--garment', required=True, help='Path to the garment image')
+    parser.add_argument('--output_garment', default='flux_inpaint_garment.png', help='Output path for garment result')
+    parser.add_argument('--output_tryon', default='flux_inpaint_tryon.png', help='Output path for try-on result')
+    parser.add_argument('--steps', type=int, default=50, help='Number of inference steps')
+    parser.add_argument('--guidance_scale', type=float, default=30, help='Guidance scale')
+    parser.add_argument('--seed', type=int, default=0, help='Random seed')
+    parser.add_argument('--width', type=int, default=576, help='Width')
+    parser.add_argument('--height', type=int, default=768, help='Height')
+    
+    args = parser.parse_args()
+    
+    check_min_version("0.30.2")
+    
+    garment_result, tryon_result = run_inference(
+        image_path=args.image,
+        mask_path=args.mask,
+        garment_path=args.garment,
+        num_steps=args.steps,
+        guidance_scale=args.guidance_scale,
+        seed=args.seed,
+        size=(args.width, args.height)
+    )
+    output_tryon_path=args.output_tryon
+    
+    tryon_result.save(output_tryon_path)
+    
+    print("Successfully saved garment and try-on images")
+
+if __name__ == "__main__":
+    main()