Lite Any Stereo V2: Faster and Stronger Efficient Zero-Shot Stereo Matching
Paper • 2606.24457 • Published • 3
This repository provides pretrained checkpoints for Lite Any Stereo V2 (LAS2), an efficient stereo matching model series designed for practical inference latency and strong zero-shot generalization.
| Model | Checkpoint |
|---|---|
| LAS2-S | LAS2_S.pth |
| LAS2-M | LAS2_M.pth |
| LAS2-L | LAS2_L.pth |
| LAS2-H | LAS2_H.pth |
The checkpoints can be downloaded programmatically with huggingface_hub:
from huggingface_hub import hf_hub_download
import torch
ckpt_path = hf_hub_download(
repo_id="tomtomtommi/LiteAnyStereoV2",
filename="LAS2_M.pth"
)
checkpoint = torch.load(ckpt_path, map_location="cpu")