LeftRefill

LeftRefill: Filling Right Canvas based on Left Reference through Generalized Text-to-Image Diffusion Model (CVPR2024)

Project Page, Paper

Abstract

This paper introduces LeftRefill, an innovative approach to efficiently harness large Text-to-Image (T2I) diffusion models for reference-guided image synthesis. As the name implies, LeftRefill horizontally stitches reference and target views together as a whole input. The reference image occupies the left side, while the target canvas is positioned on the right. Then, LeftRefill paints the right-side target canvas based on the left-side reference and specific task instructions. Such a task formulation shares some similarities with contextual inpainting, akin to the actions of a human painter.

This novel formulation efficiently learns both structural and textured correspondence between reference and target without other image encoders or adapters. We inject task and view information through cross-attention modules in T2I models, and further exhibit multi-view reference ability via the re-arranged self-attention modules. These enable LeftRefill to perform consistent generation as a generalized model without requiring test-time fine-tuning or model modifications. Thus, LeftRefill can be seen as a simple yet unified framework to address reference-guided synthesis. As an exemplar, we leverage LeftRefill to address two different challenges: reference-guided inpainting and novel view synthesis, based on the pre-trained StableDiffusion.

TODOList

• Releasing training, inference codes of Ref-inpainting and NVS (1-reference).
• Releasing pre-trained Ref-inpainting weights.
• Releasing codes of data processing.
• Releasing codes of multi-view LeftRefill.

Models

Ref-inpainting

Since our Ref-inpainting model only contains 50 trainable tokens (704KB), weights are saved in checkpoints/ref_guided_inpainting. Before the inference, you need to download SD2-inpainting model from here, and place it at ./pretrained_models/512-inpainting-ema.ckpt

Start with Gradio

You can simply use gradio to test reference-inpainting with custom masks.

python ref_inpainting_gradio.py

Training

Preprocessing from MegaDepth

Please preset data root path in megadepth_overlap.py, you could download megadepth data from link. Data index (scene_info_0.1_0.7, scene_info_val_1500) could be downloaded from LoFTR (train link, test link).

python megadepth_overlap.py

Extending Dataset for Multi-View training and testing

Please refer to extend_data_for_multiview.py, which provide an example of extending MegaDepth data for multi-view training and testing.

python extend_data_for_multiview.py

Matching mask

You could directly training LeftRefill without matching mask for a fast try, which would achieve slightly worse results. If you need matching results, please refer to some matching works for details (CasMTR, RoMa).

Reference-guided inpainting (1-reference)

CUDA_VISIBLE_DEVICES=0,1 python train_inpainting.py \
  --config_file configs/ref_inpainting_training_config.yaml \
  --exp_name RefInpainting_V0 \
  --ngpu 2 \
  --fp16

Reference-guided inpainting (multi-reference)

CUDA_VISIBLE_DEVICES=0,1 python train_inpainting.py \
  --config_file configs/multiview_ref_inpainting_training_config.yaml \
  --exp_name RefInpainting_4View_V0 \
  --ngpu 2 \
  --fp16

Novel view synthesis (1-reference)

CUDA_VISIBLE_DEVICES=0,1 python train_inpainting.py --config_file configs/nvs_training_config.yaml \
  --exp_name NVS_OBJ_V0 \
  --ngpu 2 \
  --fp16

Testing

Reference-guided inpainting

If you want to test multi-view version of reference-guided inpainting, please execute test_multiview_inpainting.py instead of test_inpainting.py.

CUDA_VISIBLE_DEVICES=0 python test_inpainting.py --model_path check_points/ref_guided_inpainting \
                                                 --batch_size 1 \
                                                 --test_path ./data/megadepth_0.4_0.7/match_test_image_pairs \ 
                                                 --cfg 2.5 \
                                                 --test_size 512 \
                                                 --metric_size 512 \
                                                 --eta 1.0 \
                                                 --output_path test_outputs_compare

Multi-View Reference-guided inpainting

CUDA_VISIBLE_DEVICES=0 python test_inpainting.py --model_path check_points/ref_guided_inpainting \
                                                 --batch_size 1 \
                                                 --test_path ./data/megadepth_0.4_0.7/match_test_image_pairs \ 
                                                 --cfg 2.5 \
                                                 --test_size 512 \
                                                 --metric_size 512 \
                                                 --eta 1.0 \
                                                 --output_path test_outputs_compare

Cite

If you found our project helpful, please consider citing:

@inproceedings{cao2024leftrefill,
      title={LeftRefill: Filling Right Canvas based on Left Reference through Generalized Text-to-Image Diffusion Model}, 
      author={Chenjie Cao and Yunuo Cai and Qiaole Dong and Yikai Wang and Yanwei Fu},
      booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
      year={2024},
}