search

ayushjain1144 / odin

Code for the paper: "ODIN: A Single Model for 2D and 3D Segmentation" (CVPR 2024)
https://odin-seg.github.io
MIT License
100 stars 8 forks source link

readme

PWC PWC

ODIN: A Single Model for 2D and 3D Segmentation

Authors: Ayush Jain, Pushkal Katara, Nikolaos Gkanatsios, Adam W. Harley, Gabriel Sarch, Kriti Aggarwal, Vishrav Chaudhary, Katerina Fragkiadaki.

Official implementation of "ODIN: A Single Model for 2D and 3D Segmentation", CVPR 2024 (Highlight).


## Installation Make sure you are using a newer version of GCC>=9.2.0 ```bash export TORCH_CUDA_ARCH_LIST="6.0 6.1 6.2 7.0 7.2 7.5 8.0 8.6" conda create -n odin python=3.10 conda activate odin pip install torch==2.2.0+cu118 torchvision==0.17.0+cu118 --extra-index-url https://download.pytorch.org/whl/cu118 pip install torch-scatter -f https://data.pyg.org/whl/torch-2.2.0+cu118.html pip install -r requirements.txt sh init.sh ``` ## Data Preparation Please refer to README in data_preparation folder for individual datasets. For eg. ScanNet data preparation [README](data_preparation/scannet/README.md) ## Usage We provide training scripts for various datasets in `scripts` folder. Please refer to these scripts for training ODIN. - Modify `DETECTRON2_DATASETS` to the path where you store the Posed RGB-D data. - Modify MODEL.WEIGHTS to load pre-trained weights. We use weights from Mask2Former. You can download the [Mask2Former-ResNet](https://huggingface.co/katefgroup/odin/resolve/main/m2f_coco.pkl) and [Mask2Former-Swin Weights](https://huggingface.co/katefgroup/odin/resolve/main/m2f_coco_swin.pkl). MODEL.WEIGHTS can also accept link to the checkpoint as well, so you can directly supply these links as the argument value. - ODIN Pre-trained weights are provided below in the Model-Zoo. Simply point to these weights using the MODEL.WEIGHTS to run inference. You would also need to add `--eval-only` flag for running evaluation. - `SOLVER.IMS_PER_BATCH` controls the batch size. This is effective batch size i.e. if you are running on 2 GPUs and the batch size is set to 6, you are using bs=3 per GPU. - `SOLVER.TEST_IMS_PER_BATCH` controls the (effective) test batch size. Since, there are variable number of images in a scene, we use bs=1 per GPU at test time. `MAX_FRAME_NUM=-1` means that it loads all images in a scene for inference, which is our usual strategy. In some datasets, the images can simply be too large, thus there we actually set a maximum limit on images. - `INPUT.SAMPLING_FRAME_NUM` controls the number of images we sample at test time -- for eg. in ScanNet, we train on 25 image chunks at training time. - `CHECKPOINT_PERIOD` is the number of iterations after which a checkpoint is saved. `EVAL_PERIOD` specifies the number of steps after which the eval is run. - `OUTPUT_DIR` stores the checkpoints and the tensorboard logs. `--resume` resumes the training from the last checkpoint stored in `OUTPUT_DIR`. If no checkpoint is present, it loads the weights from `MODEL.WEIGHTS` ## Model Zoo ScanNet Instance Segmentation | Dataset | mAP | mAP@25 | Config | Checkpoint |:-:|:-:|:-:|:-:|:-:| | ScanNet val (ResNet50) | 47.8 | 83.6 | [config](scripts/scannet/scannet_resnet.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/scannet_resnet_47.8_73.3_32k_1.5k.pth) | ScanNet val (Swin-B) | 50.0 | 83.6 | [config](scripts/scannet/scannet_swin.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/scannet_swin_50.0_64k_6k.pth) | ScanNet test (Swin-B) | 47.7 | 86.2 | [config](scripts/scannet/scannet_swin_benchmark.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/scannet_swin_50.0_64k_6k.pth) ScanNet Semantic Segmentation | Dataset | mIoU | Config | Checkpoint |:-:|:-:|:-:|:-:| | ScanNet val (ResNet50) | 73.3 | [config](scripts/scannet/scannet_resnet.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/scannet_resnet_47.8_73.3_32k_1.5k.pth) | ScanNet val (Swin-B) | 77.8 | [config](scripts/scannet/scannet_swin.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/scannet_swin_semantic_77.8_64k_2k.pth) | ScanNet test (Swin-B) | 74.4 | [config](scripts/scannet/scannet_benchmark.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/scannet_swin_semantic_77.8_64k_2k.pth) Joint 2D-3D on ScanNet and COCO | Model | mAP (ScanNet) | mAP25 (ScanNet) | mAP (COCO) | Config | Checkpoint |:-:|:-:|:-:|:-:|:-:|:-:| | ODIN | 49.1 | 83.1 | 41.2 | [config](scripts/joint_coco_scannet/scannet_coco_resnet.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/scannet_coco_joint_49.1_41.2_88k_19k.pth) ScanNet200 Instance Segmentation | Dataset | mAP | mAP@25 | Config | Checkpoint |:-:|:-:|:-:|:-:|:-:| | ScanNet200 val (ResNet50) | 25.6 | 36.9 | [config](scripts/scannet200/scannet200_resnet.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/scannet200_resnet_25.6_35.8_52k_5k.pth) | ScanNet200 val (Swin-B) | 31.5 | 45.3 | [config](scripts/scannet200/scannet200_swin.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/scannet200_swin_31.5_76k_5.5k.pth) | ScanNet200 test (Swin-B) | 27.2 | 39.4 | [config](scripts/scannet200/scannet200_swin_benchmark.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/scannet200_swin_31.5_76k_5.5k.pth) ScanNet200 Semantic Segmentation | Dataset | mIoU | Config | Checkpoint |:-:|:-:|:-:|:-:| | ScanNet200 val (ResNet50) | 35.8 | [config](scripts/scannet200/scannet200_resnet.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/scannet200_resnet_25.6_35.8_52k_5k.pth) | ScanNet200 val (Swin-B) | 40.5 | [config](scripts/scannet/scannet_val.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/scannet200_swin_40.5_76k_3k.pth) | ScanNet test (Swin-B) | 36.8 | [config](scripts/scannet/scannet_benchmark.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/scannet200_swin_40.5_76k_3k.pth) AI2THOR Semantic and Instance Segmentation | Dataset | mAP | mAP@25 | mIoU | Config | Checkpoint |:-:|:-:|:-|:-:|:-:|:-:| | AI2THOR val (ResNet) | 63.8 | 80.2 | 71.5 | [config](scripts/ai2thor/ai2thor_resnet.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/ai2thor_resnet_63.8_71.4_1.5k.pth) | AI2RHOR val (Swin) | 64.3 | 78.6 | 71.4 | [config](scripts/ai2thor/ai2thor_swin.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/ai2thor_swin_64.3_71.4_112k_4.5k.pth) Matterport3D Instance Segmentation | Dataset | mAP | mAP@25 | Config | Checkpoint |:-:|:-:|:-|:-:|:-:| | Matterport3D val (ResNet) | 11.5 | 27.6 | [config](scripts/matterport3d/matterport_resnet.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/matterport_resnet_11.5_22.4_84k_13k.pth) | Matterport val (Swin) | 14.5 | 36.8 | [config](scripts/matterport3d/matterport_swin.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/matterport_swin_14.5_68k_16k.pth) Matterport3D Semantic Segmentation | Dataset | mIoU | mAcc | Config | Checkpoint |:-:|:-:|:-:|:-:|:-:| | Matterport3D val (ResNet) | 22.4 | 28.5 | [config](scripts/matterport3d/matterport_resenet.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/matterport_resnet_11.5_22.4_84k_13k.pth) | Matterport3D val (Swin) | 28.6 | 38.2 | [config](scripts/matterport3d/matterport_swin.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/matterport_swin_28.6_68k_20k.pth) S3DIS Instance Segmentation | Dataset | mAP | mAP@25 | Config | Checkpoint |:-:|:-:|:-:|:-:|:-:| | S3DIS Area5 (ResNet50-Scratch) | 36.3 | 61.2 | [config](scripts/s3dis/s3dis_resnet_scratch.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/s3dis_resnet_36.3_59.7_24k_0.5k.pth) | S3DIS Area5 (ResNet50-Fine-Tuned) | 44.7 | 67.5 | [config](scripts/s3dis/s3dis_resnet_scratch.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/s3dis_resnet_finetuned_44.7_66.7_6.5k_1.5k.pth) | S3DIS Area5 (Swin-B) | 43.0 | 70.0 | [config](scripts/s3dis/s3dis_swin_finetuned.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/s3dis_swin_finetuned_43.0_68.6_4.5k_3k.pth) S3DIS Semantic Segmentation | Dataset | mIoU | Config | Checkpoint |:-:|:-:|:-:|:-:| | S3DIS (ResNet50) | 59.7 | [config](scripts/s3dis/s3dis_resnet_finetuned.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/s3dis_resnet_36.3_59.7_24k_0.5k.pth) | Swin-B | 68.6 | [config](scripts/s3dis_swin_finetuned.sh) | [checkpoint](https://huggingface.co/katefgroup/odin/resolve/main/s3dis_swin_finetuned_43.0_68.6_4.5k_3k.pth) ## Training Logs: Please find training logs for all models [here](https://huggingface.co/katefgroup/odin_logs/tensorboard) ## Citing ODIN If you find ODIN useful in your research, please consider citing: ```BibTeX @misc{jain2024odin, title={ODIN: A Single Model for 2D and 3D Perception}, author={Ayush Jain and Pushkal Katara and Nikolaos Gkanatsios and Adam W. Harley and Gabriel Sarch and Kriti Aggarwal and Vishrav Chaudhary and Katerina Fragkiadaki}, year={2024}, eprint={2401.02416}, archivePrefix={arXiv}, primaryClass={cs.CV} } ``` ## License Shield: [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT) The majority of ODIN is licensed under a [MIT License](LICENSE). ## Acknowledgement Parts of this code were based on the codebase of [Mask2Former](https://github.com/facebookresearch/Mask2Former) and [Mask3D](https://github.com/JonasSchult/Mask3D).