- TSD, PC
- Deformable roi pooling(https://github.com/open-mmlab/mmdetection)
- Backbone Faster-RCNN-FPN-50
- Set hyper variable as paper does
(except base_lr and batch because of gpu)
-
TSD, PC
a. Add TSDHeads instead ROIHeads(Merged)
b. Add Deformation to disentangle regression and classification in TSDHead
c. Add DeformableROIPooler for classification in TSDHead
d. Add FastRCNNRegressionOutputLayers, FastRCNNClassificationOutputLayers in TSDHead for prediction
e. Add progressive constraints on revisit.py -
Deformable ROI Pooling
a. Using Deformable ROI Pooling code for classification implemented at /detectron2/layers/csrc b. Take original ROI feature as input c. First layer is shared with regression -
Hyperparameters
a. M_c = M_r = 0.2
b. pooling size = 7
c. gamma = 0.1
VOC 2007 + VOC 2012
- Training set: 16551 images
- Validation set: 4952 images (VOC 2007 only)
Ubuntu 18.04
Anaconda 4.7.5
Pytorch 1.4.0
Python 3.8.2
CUDA 10.0
CUDNN 7.6.3
See INSTALL.md.
See GETTING_STARTED.md, or the Colab Notebook.
Learn more at our documentation. And see projects/ for some projects that are built on top of detectron2.
We provide a script in "tools/revisit_train_net.py", that is made to train all the configs provided in revisit. You may want to use it as a reference to write your own training script for a new research.
To train a model with "revisit_train_net.py", first setup the corresponding datasets following datasets/README.md, then run:
cd tools/
./revisit_train_net.py --num-gpus 8 \
--config-file ../configs/PascalVOC-Detection/faster_rcnn_R_50_FPN.yaml
The configs are made for 8-GPU training. To train on 1 GPU, you may need to change some parameters, e.g.:
./revisit_train_net.py \
--config-file ../configs/PascalVOC-Detection/faster_rcnn_R_50_FPN_1GPU.yaml \
SOLVER.IMS_PER_BATCH 2 SOLVER.BASE_LR 0.0025
For most models, CPU training is not supported.
To evaluate a model's performance, use
./revisit_train_net.py \
--config-file ../configs/PascalVOC-Detection/faster_rcnn_R_50_FPN.yaml \
--eval-only MODEL.WEIGHTS /path/to/checkpoint_file
For more options, see ./revisit_train_net.py -h
.
Detectron2 is released under the Apache 2.0 license.
If you use Detectron2 in your research or wish to refer to the baseline results published in the Model Zoo, please use the following BibTeX entry.
@misc{wu2019detectron2,
author = {Yuxin Wu and Alexander Kirillov and Francisco Massa and
Wan-Yen Lo and Ross Girshick},
title = {Detectron2},
howpublished = {\url{https://github.com/facebookresearch/detectron2}},
year = {2019}
}