ResNet-50, from scratch trained on NYUv2

Demo: predicted vs. ground-truth depth on NYUv2 test scenes from our dataset mirror (per-frame AbsRel, Eigen crop), rendered with the official Monocular-Depth-Estimation-Toolbox code.

Unofficial mirror of the official SimIPU + depth head (Monocular-Depth-Estimation-Toolbox) weights from zhyever/Monocular-Depth-Estimation-Toolbox. The weights are unchanged; this repository adds a weights-only model.safetensors copy and a config.json.

Model

ResNet-50 encoder with the toolbox's depth head, fine-tuned on NYUv2 for 24 epochs from different encoder initialisations (none, ImageNet supervised, SimIPU on KITTI, SimIPU on Waymo).

Results (reported by the authors)

Metric Value
Abs Rel (NYUv2, +flip) 0.205

See the official repository for full tables and evaluation details.

Files

File Content
scratch_nyu.pth original checkpoint, exactly as released
model.safetensors the model weights only (bit-exact copy of the checkpoint's state_dict tensors; no optimizer state)
config.json architecture, dataset, results and provenance

Usage

import json
from huggingface_hub import hf_hub_download

repo = "dronefreak/simipu-r50-nyuv2-scratch"
config = json.load(open(hf_hub_download(repo, "config.json")))
ckpt = hf_hub_download(repo, "scratch_nyu.pth")   # original format: use it with the official code
print(config["architecture"], config["dataset"], ckpt)

Build the network with the official code (https://github.com/zhyever/Monocular-Depth-Estimation-Toolbox) using the matching config for this variant, then load scratch_nyu.pth 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:

@article{li2021simipu,
  title   = {SimIPU: Simple 2D Image and 3D Point Cloud Unsupervised Pre-Training for Spatial-Aware Visual Representations},
  author  = {Li, Zhenyu and Chen, Zehui and Li, Ang and Fang, Liangji and Jiang, Qinhong and Liu, Xianming and Jiang, Junjun and Zhou, Bolei and Zhao, Hang},
  journal = {arXiv preprint arXiv:2112.04680},
  year    = {2021}
}
@misc{lidepthtoolbox2022,
  title        = {Monocular Depth Estimation Toolbox},
  author       = {Zhenyu Li},
  howpublished = {\url{https://github.com/zhyever/Monocular-Depth-Estimation-Toolbox}},
  year         = {2022}
}

Disclaimer

This is not an official release. All credit goes to the original authors (Zhenyu Li et al.; Monocular-Depth-Estimation-Toolbox). The weights are redistributed under the Apache-2.0 license of the official repository. If you are an author and want this mirror changed or removed, please open a discussion.

Downloads last month
-
Safetensors
Model size
77.8M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Dataset used to train dronefreak/simipu-r50-nyuv2-scratch

Collection including dronefreak/simipu-r50-nyuv2-scratch

Paper for dronefreak/simipu-r50-nyuv2-scratch