Many video editing tasks such as rotoscoping or object removal require the propagation of context across frames. While transformers and other attention-based approaches that aggregate features globally have demonstrated great success at propagating object masks from keyframes to the whole video, they struggle to propagate high-frequency details such as textures faithfully. We hypothesize that this is due to an inherent bias of global attention towards low-frequency features. To overcome this limitation, we present a two-stream approach, where high-frequency features interact locally and low-frequency features interact globally. The global interaction stream remains robust in difficult situations such as large camera motions, where explicit alignment fails. The local interaction stream propagates high-frequency details through deformable feature aggregation and, informed by the global interaction stream, learns to detect and correct errors of the deformation field. We evaluate our two-stream approach for inpainting tasks, where experiments show that it improves both the propagation of features within a single frame as required for image inpainting, as well as their propagation from keyframes to target frames. Applied to video inpainting, our approach leads to 44% and 26% improvements in FID and LPIPS scores.
Towards Unified Keyframe Propagation Models
Patrick Esser, Peter Michael, Soumyadip Sengupta
comparison_small.mp4
Video results for all DAVIS sequences.
conda env create -f env.yaml
conda activate guided-inpainting
Download the raft-things.pth checkpoint for
RAFT and place into
checkpoints/flow/raft/raft-things.pth.
Download the encoder_epch_20.pth checkpoint for the
ade20k-resnet50dilated-ppm_deepsup perceptual loss of
LaMa and place into
checkpoints/lama/ade20k/ade20k-resnet50dilated-ppm_deepsup/encoder_epoch_20.pth.
To reproduce the results from Table 2, download the validation data
to data/places365/lama/val_guided/. Download the desired pre-trained
checkpoint(s) and run
python gi/main.py --base configs/<model>.yaml --gpus 0, --train false --resume_from_checkpoint models/<model>.ckpt
To reproduce the results in Table 3, set up the DEVIL benchmark, download the pre-computed results and run DEVIL on it. Note that at the moment we do not plan to release the propagation code.
Follow the training split preparation of Places as in
LaMa and place into
data/places365/data_large. Start the training with
python gi/main.py --base configs/<model>.yaml --gpus 0,1