Skip to content

Latest commit

 

History

History
51 lines (44 loc) · 1.54 KB

README.md

File metadata and controls

51 lines (44 loc) · 1.54 KB
Raw

Object detection using adaptive sampling

Environment

CUDA 11.1
pytorch 1.9.0

The following is required to be installed before running:

torvision 0.10.0
pyyaml
scipy
matplotlib
opencv-python
tensorboardX

Run the following commands to install pointnet2 and weighted FPS:

cd external
cd pointnet2
python setup.py install
cd ..
cd weighted_FPS
python setup.py install

Data preparation

Download TOS dataset with bbox, instance label and semantic label. Then, modify the dataset_dir and save_dir under ./utils/generate_gt_heatmap.py. Then run this script to generate the ground truth objectness map that is mentioned in the paper.
Then, put the TO-xxx-wHM folder under ./data

Usage

  • Train

For training the TO-crowd or TO-vanilla, it requires pretraining the heatmap module firstly, to train the heatmap module, run the following commands:

python main.py --mode train --config ./configs/train_heatmap.yaml

Then, modify the hm_pretrain_path in ./configs/train_vote_adaptive_desk.yaml to point to the pretrain heatmap module's weight.
Finally run the adaptive votenet by:

python main.py --mode train --config ./configs/train_vote_adaptive_desk.yaml

Original VoteNet baseline

If you wish to run original VoteNet on our dataset, you may refer to the usage on their official repo and our implementation here.