QueryDet-PyTorch
This repository is the official implementation of our paper: QueryDet: Cascaded Sparse Query for Accelerating High-Resolution Small Object Detection
Requirement
a. Install Pytorch 1.4 following here
b. Install APEX following here
c. Install our Pytorch based sparse convolution operation following here
d. Install the detectron2 toolkit following here, note that we build our approach based on version 0.2.1. Note you may follow the instructions to set COCO configs
d. Clone our repository and have fun with it!
Usage
1. Data preparation
a. To prepare MS-COCO, you may follow the instructions of Detectron2
b. We provide the data preprocessing code for VisDrone2018. You need to first download dataset from here
c. Check visdrone/data_prepare.py to process the dataset
2. Training
% train coco RetinaNet baseline
python train_coco.py --config-file models/retinanet/configs/coco/train.yaml --num-gpu 8 OUTPUT_DIR /path/to/workdir
% train coco QueryDet
python train_coco.py --config-file models/querydet/configs/coco/train.yaml --num-gpu 8 OUTPUT_DIR /path/to/workdir
% train VisDrone RetinaNet baseline
python train_visdrone.py --config-file models/retinanet/configs/visdrone/train.yaml --num-gpu 8 OUTPUT_DIR /path/to/workdir
% train VisDrone QueryDet
python train_visdrone.py --config-file models/querydet/configs/visdrone/train.yaml --num-gpu 8 OUTPUT_DIR /path/to/workdir
3. Test
% test coco RetinaNet baseline
python infer_coco.py --config-file models/retinanet/configs/coco/test.yaml --num-gpu 8 --eval-only MODEL.WEIGHTS /path/to/workdir/model_final.pth
% test coco QueryDet with Dense Inference
python infer_coco.py --config-file models/querydet/configs/coco/test.yaml --num-gpu 8 --eval-only MODEL.WEIGHTS /path/to/workdir/model_final.pth
% test coco QueryDet with CSQ
python infer_coco.py --config-file models/querydet/configs/coco/test.yaml --num-gpu 8 --eval-only MODEL.WEIGHTS /path/to/workdir/model_final.pth MODEL.QUERY.QUERY_INFER True
54 Nov 21, 2022
5 Sep 16, 2022
16 Oct 26, 2022
17 May 06, 2022
51 Dec 29, 2022
75 Dec 02, 2022
7.8k Dec 30, 2022
65 Dec 21, 2022
5 Feb 08, 2022
46 Sep 29, 2022
13 Dec 13, 2022
573 Jan 04, 2023
34 Nov 02, 2022
30 Dec 12, 2022
51 Nov 18, 2022
1 Oct 22, 2021
1.8k Dec 30, 2022
57 Dec 27, 2022
23 Sep 10, 2022
9 Dec 13, 2021