NeWCRFs (Swin-L) trained on KITTI (Eigen split)
Demo: predicted depth on consecutive frames of KITTI raw drive 2011_09_26_drive_0002 (an Eigen test drive). KITTI images © Geiger et al., KIT & Toyota Technological Institute, CC BY-NC-SA 3.0 (https://www.cvlibs.net/datasets/kitti/); rendered with the official code.
Unofficial mirror of the official NeWCRFs weights from aliyun/NeWCRFs. The weights are unchanged; this repository adds a weights-only
model.safetensorscopy and aconfig.json.
Model
- Architecture: NeWCRFs: NeWCRFs (Swin-L)
- Task: monocular depth estimation
- Trained on: KITTI (Eigen split)
- Parameters: 270.4M
- Paper: NeW CRFs: Neural Window Fully-connected CRFs for Monocular Depth Estimation (CVPR 2022)
- Code: https://github.com/aliyun/NeWCRFs
- License: MIT (from the official repository)
Swin Transformer encoder with neural-window fully-connected CRF decoder.
Results (reported by the authors)
| Metric | Value |
|---|---|
| Abs Rel | 0.052 |
| Sq Rel | 0.1482 |
| RMSE | 2.0716 |
| RMSE log | 0.078 |
| delta<1.25 | 0.975 |
| delta<1.25^2 | 0.997 |
| delta<1.25^3 | 0.999 |
| SILog | 6.9859 |
See the official repository for full tables and evaluation details.
Files
| File | Content |
|---|---|
model_kittieigen.ckpt |
original checkpoint, exactly as released |
model.safetensors |
the model weights only (bit-exact copy of the checkpoint's model tensors; no optimizer state) |
config.json |
architecture, dataset, results and provenance |
Usage
import json
from huggingface_hub import hf_hub_download
repo = "dronefreak/newcrfs-kitti"
config = json.load(open(hf_hub_download(repo, "config.json")))
ckpt = hf_hub_download(repo, "model_kittieigen.ckpt") # original format: use it with the official code
print(config["architecture"], config["dataset"], ckpt)
Build the network with the official code (https://github.com/aliyun/NeWCRFs) using the matching config for this variant, then load
model_kittieigen.ckpt exactly as you would the original download, or load model.safetensors with
safetensors.torch.load_file into the same model.
Citation
If you use these weights, please cite the original work:
@inproceedings{yuan2022newcrfs,
title = {NeWCRFs: Neural Window Fully-connected CRFs for Monocular Depth Estimation},
author = {Yuan, Weihao and Gu, Xiaodong and Dai, Zuozhuo and Zhu, Siyu and Tan, Ping},
booktitle = {CVPR},
year = {2022}
}
Disclaimer
This is not an official release. All credit goes to the original authors (Weihao Yuan, Xiaodong Gu, Zuozhuo Dai, Siyu Zhu, Ping Tan; Alibaba Group). The weights are redistributed under the MIT license of the official repository. If you are an author and want this mirror changed or removed, please open a discussion.
- Downloads last month
- -