ImVoteNet stage 2 (PointNet++) trained on SUN RGB-D
Demo: 3D detections (score ≥ 0.4) of this checkpoint on SUN RGB-D test scenes from our dataset mirror: point cloud top view and boxes projected onto the image, rendered with the official MMDetection3D code.
Unofficial mirror of the official ImVoteNet (MMDetection3D) weights from open-mmlab/mmdetection3d. The weights are unchanged; this repository adds a weights-only
model.safetensorscopy and aconfig.json.
Model
- Architecture: ImVoteNet (MMDetection3D): ImVoteNet stage 2 (PointNet++)
- Task: 3D object detection from RGB-D (point cloud + image)
- Trained on: SUN RGB-D (mirror on the Hub)
- Parameters: 43.5M
- Paper: ImVoteNet: Boosting 3D Object Detection in Point Clouds with Image Votes (CVPR 2020)
- Code: https://github.com/open-mmlab/mmdetection3d
- License: Apache-2.0 (from the official repository)
Stage 2 (3D detector with PointNet++ backbone and image votes), 10 SUN RGB-D classes. The stage-1 image branch (Faster R-CNN R50-FPN on SUN RGB-D 2D) that stage 2 builds on is included under pretrained/. The original facebookresearch/imvotenet repo did not release trained checkpoints; these are the official MMDetection3D checkpoints (v1.0.0_models).
Results (reported by the authors)
| Metric | Value |
|---|---|
| AP@0.25 (SUN RGB-D 3D, 10 classes) | 64.48 |
See the official repository for full tables and evaluation details.
Files
| File | Content |
|---|---|
imvotenet_stage2_16x8_sunrgbd-3d-10class_20210819_192851-1bcd1b97.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 |
pretrained/imvotenet_faster_rcnn_r50_fpn_2x4_sunrgbd-3d-10class_20210819_225618-62eba6ce.pth |
stage 1: image branch (Faster R-CNN R50-FPN) on SUN RGB-D 2D, AP@0.5 62.70 |
Usage
import json
from huggingface_hub import hf_hub_download
repo = "dronefreak/imvotenet-sunrgbd"
config = json.load(open(hf_hub_download(repo, "config.json")))
ckpt = hf_hub_download(repo, "imvotenet_stage2_16x8_sunrgbd-3d-10class_20210819_192851-1bcd1b97.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/open-mmlab/mmdetection3d) using the matching config for this variant, then load
imvotenet_stage2_16x8_sunrgbd-3d-10class_20210819_192851-1bcd1b97.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:
@inproceedings{qi2020imvotenet,
title = {Imvotenet: Boosting 3D object detection in point clouds with image votes},
author = {Qi, Charles R and Chen, Xinlei and Litany, Or and Guibas, Leonidas J},
booktitle = {CVPR},
pages = {4404--4413},
year = {2020}
}
@misc{mmdet3d2020,
title = {{MMDetection3D: OpenMMLab} next-generation platform for general {3D} object detection},
author = {MMDetection3D Contributors},
howpublished = {\url{https://github.com/open-mmlab/mmdetection3d}},
year = {2020}
}
Disclaimer
This is not an official release. All credit goes to the original authors (Charles R. Qi, Xinlei Chen, Or Litany, Leonidas J. Guibas (method); OpenMMLab (checkpoint); OpenMMLab (MMDetection3D model zoo); method by Facebook AI Research). 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
- -