Skip to content

Latest commit

 

History

History
87 lines (63 loc) · 2.68 KB

README.md

File metadata and controls

87 lines (63 loc) · 2.68 KB
Raw

Prepare ScanNet-SV data for training monocular RGB-D perception model

Step 1. Prepare ScanNet 2D data. The processed 2D data can be downloaded from HERE. Run cat 2D.tar.* > 2D.tar to merge the files. Then skip to Step 2.

Or you can process ScanNet 2D data yourself by following the steps below.

First acquire download-scannet.py from HERE. You should fill out an agreement to the ScanNet Terms of Use.

Then download 2D instance data, run:

python download-scannet.py -o <ScanNet root> --type _2d-instance.zip

Extract _2d-instance.zip into 2D_info folder, whose structure follows:

2D_info
└── scenexxxx_xx_2d-instance/instance/xxxxxx.png

Link the 2D_info folder to this level of directory.

Process 2D instance data by:

python prepare_2d_ins.py --scannet_path ./2D_info --output_path ./2D --scene_index_file ./meta_data/scannet_train.txt

Step 2. Prepare scannet_frames_25k data, run:

python download-scannet.py -o <ScanNet root> --preprocessed_frames

Link or move the scannet_frames_25k folder to this level of directory.

Step 3. Prepare ScanNet 3D data. Follow votenet to download and process the 3D data. Link or move the scans folder to this level of directory.

Process SV data by running python generate_18cls_gt_data.py, which will create two folders named scannet_sv_18cls_train and scannet_sv_18cls_val here.

Step 4. Generate .pkl files by:

python tools/create_data.py scannet --root-path ./data/scannet-sv --out-dir ./data/scannet-sv --extra-tag scannet_sv

Final folder structure:

scannet-sv
├── README.md
├── scannet_frames_25k/
├── meta_data/
│   ├── scannet_train.txt
│   ├── scannetv2_train.txt
│   ├── scannetv2_val.txt
├── scans/
├── 2D_info/
├── 2D
│   ├── scenexxxx_xx
│   │   ├── instance
│   │   │   ├── xxxxxx.png
├── scannet_sv_18cls_train/
├── scannet_sv_18cls_val/
│   ├── scenexxxx_xx_xxxxxx_bbox.npy
│   ├── scenexxxx_xx_xxxxxx_ins_label.npy
│   ├── scenexxxx_xx_xxxxxx_pc.npy
│   ├── scenexxxx_xx_xxxxxx_pose.txt
│   ├── scenexxxx_xx_xxxxxx_sem_label.npy
│   ├── scenexxxx_xx_xxxxxx.jpg
│   ├── scenexxxx_xx_depth_intrinsic.txt
│   ├── scenexxxx_xx_image_intrinsic.txt
├── prepare_2d_ins.py
├── generate_18cls_gt_data.py
├── load_scannet_data.py
├── scannet_utils.py
├── scannet_sv_infos_train.pkl
└── scannet_sv_infos_val.pkl