v1.2 — Perceptual + temporal fine-tune

dabit_perceptual.pth — the v1.1 dabit_retrained.pth model fine-tuned for 50k iterations with an added LPIPS-VGG perceptual loss and a flow-warped, occlusion-masked temporal-consistency loss (Lai et al., ECCV 2018). Config: configs/dabit_ft.json, losses.perc_weight / losses.temp_weight.

Result (all 90 DAVIS-Blur sequences, fp32)

Model	PSNR ↑	SSIM ↑	LPIPS ↓	tOF ↓
v1.1 retrain	29.17	0.858	0.217	1.103
v1.2 perceptual	28.99	0.856	0.173	0.882

Trades ~0.18 dB PSNR for 20% lower LPIPS (sharper perceptual detail) and 20% lower tOF (smoother motion). Choose this model for perceptual quality, the v1.1 retrain for peak PSNR/SSIM.

Install

Requires the RAFT + Depth Anything V2 dependency weights from v1.1:

mkdir -p weights
gh release download v1.2 -R crispianm/DaBiT -D weights   # perceptual model
gh release download v1.1 -R crispianm/DaBiT -D weights   # deps (raft, depth-anything)
python test_dabit.py --model weights/dabit_perceptual.pth

v1.1 — Retrained model + all weights

Every weight needed to run DaBiT (training, evaluation, or inference), plus a retrained model that improves on the paper — no architectural change, purely more and more diverse clean training video.

Models

Model	PSNR ↑	SSIM ↑	LPIPS ↓	tOF ↓
dabit.pth — original paper model	28.48	0.841	0.242	1.189
dabit_retrained.pth — retrained, expanded data	29.17	0.858	0.217	1.103

Mean over all 90 DAVIS-Blur sequences, full fp32 inference.

dabit_retrained.pth uses the paper architecture and recipe, retrained for 300k iterations on a ~1.4M-frame clean-video corpus (YouTube-VOS + BVI-DVC + TartanAir-V2 + Virtual KITTI 2). +0.69 dB PSNR over the released paper weights. For the perceptually-tuned variant, see v1.2.

Dependency weights (required to run either model)

• raft.pth — RAFT optical flow (redistributed unchanged from princeton-vl/RAFT, BSD-3)
• depth_anything_v2_vits.pth — used at train/test time
• depth_anything_v2_vitb.pth, depth_anything_v2_vitl.pth — for get_depths.py (Depth Anything V2, Apache-2.0)

Install

mkdir -p weights && gh release download v1.1 -R crispianm/DaBiT -D weights
python test_dabit.py --model weights/dabit_retrained.pth