The major contributors of this repository include Yuwen Xiong, Haozhi Qi, Guodong Zhang, Yi Li, Jifeng Dai, Bin Xiao, Han Hu and Yichen Wei.
Deformable ConvNets is initially described in an arxiv tech report.
R-FCN is initially described in a NIPS 2016 paper.
This is an official implementation for Deformable Convolutional Networks (Deformable ConvNets) based on MXNet. It is worth noticing that:
- The original implementation is based on our internal Caffe version on Windows. There are slight differences in the final accuracy and running time due to the plenty details in platform switch.
- The code is tested on official MXNet@(commit 62ecb60) with the extra operators for Deformable ConvNets.
- We trained our model based on the ImageNet pre-trained ResNet-v1-101 using a model converter. The converted model produces slightly lower accuracy (Top-1 Error on ImageNet val: 24.0% v.s. 23.6%).
- This repository used code from MXNet rcnn example and mx-rfcn.
© Microsoft, 2017. Licensed under an Apache-2.0 license.
If you find Deformable ConvNets useful in your research, please consider citing:
@article{dai17dcn,
Author = {Jifeng Dai, Haozhi Qi, Yuwen Xiong, Yi Li, Guodong Zhang, Han Hu, Yichen Wei},
Title = {Deformable Convolutional Networks},
Journal = {arXiv preprint arXiv:1703.06211},
Year = {2017}
}
@inproceedings{dai16rfcn,
Author = {Jifeng Dai, Yi Li, Kaiming He, Jian Sun},
Title = {{R-FCN}: Object Detection via Region-based Fully Convolutional Networks},
Conference = {NIPS},
Year = {2016}
}
training data | testing data | [email protected] | [email protected] | time | |
---|---|---|---|---|---|
R-FCN, ResNet-v1-101 | VOC 07+12 trainval | VOC 07 test | 79.6 | 63.1 | 0.16s |
Deformable R-FCN, ResNet-v1-101 | VOC 07+12 trainval | VOC 07 test | 82.3 | 67.8 | 0.19s |
training data | testing data | mAP | [email protected] | [email protected] | mAP@S | mAP@M | mAP@L | |
---|---|---|---|---|---|---|---|---|
R-FCN, ResNet-v1-101 | coco trainval | coco test-dev | 32.1 | 54.3 | 33.8 | 12.8 | 34.9 | 46.1 |
Deformable R-FCN, ResNet-v1-101 | coco trainval | coco test-dev | 35.7 | 56.8 | 38.3 | 15.2 | 38.8 | 51.5 |
Faster R-CNN, ResNet-v1-101 | coco trainval | coco test-dev | 30.3 | 52.1 | 31.4 | 9.9 | 32.2 | 47.4 |
Deformable Faster R-CNN, ResNet-v1-101 |
coco trainval | coco test-dev | 35.0 | 55.0 | 38.3 | 14.3 | 37.7 | 52.0 |
training data | testing data | mIoU | time | |
---|---|---|---|---|
DeepLab, ResNet-v1-101 | Cityscapes train | Cityscapes val | 70.3 | 0.51s |
Deformable DeepLab, ResNet-v1-101 | Cityscapes train | Cityscapes val | 75.2 | 0.52s |
DeepLab, ResNet-v1-101 | VOC 12 train (augmented) | VOC 12 val | 70.7 | 0.08s |
Deformable DeepLab, ResNet-v1-101 | VOC 12 train (augmented) | VOC 12 val | 75.9 | 0.08s |
Running time is counted on a single Maxwell Titan X GPU (mini-batch size is 1 in inference).
-
MXNet from the offical repository. We tested our code on MXNet@(commit 62ecb60). Due to the rapid development of MXNet, it is recommended to checkout this version if you encounter any issues. We may maintain this repository periodically if MXNet adds important feature in future release.
-
Python 2.7. We recommend using Anaconda2
-
Python packages might missing: cython, opencv-python >= 3.2.0, easydict. If
pip
is set up on your system, those packages should be able to be fetched and installed by runningpip install Cython pip install opencv-python==3.2.0.6 pip install easydict==1.6
-
For Windows users, Visual Studio 2015 is needed to compile cython module.
Any NVIDIA GPUs with at least 4GB memory should be OK.
- Clone the Deformable ConvNets repository
git clone https://github.com/msracver/Deformable-ConvNets.git
- For Windows users, run
cmd .\init.bat
. For Linux user, runsh ./init.sh
. The scripts will build cython module automatically and create some folders. - Copy operators in
./rfcn/operator_cxx
to$(YOUR_MXNET_FOLDER)/src/operator/contrib
and recompile MXNet. - Please install MXNet following the official guide of MXNet. For advanced users, you may put your Python packge into
./external/mxnet/$(YOUR_MXNET_PACKAGE)
, and modifyMXNET_VERSION
in./experiments/rfcn/cfgs/*.yaml
to$(YOUR_MXNET_PACKAGE)
. Thus you can switch among different versions of MXNet quickly. - For Deeplab, we use the argumented VOC 2012 dataset. The argumented annotations are provided by SBD dataset. For convenience, we provide the converted PNG annotations and the lists of train/val images, please download them from OneDrive.
-
To use the demo with our trained model (on COCO trainval), please download the model manually from OneDrive, and put it under folder
model/
.Make sure it looks like this:
./model/rfcn_dcn_coco-0000.params ./model/rfcn_coco-0000.params ./model/deeplab_dcn_cityscapes-0000.params ./model/deeplab_cityscapes-0000.params ./model/deform_conv-0000.params ./model/deform_psroi-0000.params
-
To run the R-FCN demo, run
python ./rfcn/demo.py
By default it will run Deformable R-FCN and gives several prediction results, to run R-FCN, use
python ./rfcn/demo.py --rfcn_only
-
To run the DeepLab demo, run
python ./deeplab/demo.py
By default it will run Deformable Deeplab and gives several prediction results, to run DeepLab, use
python ./deeplab/demo.py --deeplab_only
-
To visualize the offset of deformable convolution and deformable psroipooling, run
python ./rfcn/deform_conv_demo.py python ./rfcn/defrom_psroi_demo.py
For R-FCN:
-
Please download COCO and VOC 2007+2012 datasets, and make sure it looks like this:
./data/coco/ ./data/VOCdevkit/VOC2007/ ./data/VOCdevkit/VOC2012/
-
Please download ImageNet-pretrained ResNet-v1-101 model manually from OneDrive, and put it under folder
./model
. Make sure it looks like this:./model/pretrained_model/resnet_v1_101-0000.params
For DeepLab:
-
Please download Cityscapes and VOC 2012 datasets and make sure it looks like this:
./data/cityscapes/ ./data/VOCdevkit/VOC2012/
-
Please download argumented VOC 2012 annotations/image lists, and put the argumented annotations and the argumented train/val lists into:
./data/VOCdevkit/VOC2012/SegmentationClass/ ./data/VOCdevkit/VOC2012/ImageSets/Main/
, Respectively.
-
Please download ImageNet-pretrained ResNet-v1-101 model manually from OneDrive, and put it under folder
./model
. Make sure it looks like this:./model/pretrained_model/resnet_v1_101-0000.params
-
All of our experiment settings (GPU #, dataset, etc.) are kept in yaml config files at folder
./experiments/rfcn/cfgs
and./experiments/deeplab/cfgs/
. -
Eight config files have been provided so far, namely, R-FCN for COCO/VOC, Deformable R-FCN for COCO/VOC, Deeplab for Cityscapes/VOC and Deformable Deeplab for Cityscapes/VOC, respectively. We use 8 and 4 GPUs to train models on COCO and on VOC for R-FCN, respectively. For deeplab, we use 4 GPUs for all experiments.
-
To perform experiments, run the python scripts with the corresponding config file as input. For example, to train and test deformable convnets on COCO with ResNet-v1-101, use the following command
python experiments\rfcn\rfcn_end2end_train_test.py --cfg experiments\rfcn\cfgs\resnet_v1_101_coco_trainval_rfcn_dcn_end2end_ohem.yaml
A cache folder would be created automatically to save the model and the log under
output/rfcn_dcn_coco/
. -
Please find more details in config files and in our code.
Code has been tested under:
- Ubuntu 14.04 with a Maxwell Titan X GPU and Intel Xeon CPU E5-2620 v2 @ 2.10GHz
- Windows Server 2012 R2 with 8 K40 GPUs and Intel Xeon CPU E5-2650 v2 @ 2.60GHz
- Windows Server 2012 R2 with 4 Pascal Titan X GPUs and Intel Xeon CPU E5-2650 v4 @ 2.30GHz
Q: It says AttributeError: 'module' object has no attribute 'DeformableConvolution'
.
A: This is because either
- you forget to copy the operators to your MXNet folder
- or you copy to the wrong path
- or you forget to re-compile
- or you install the wrong MXNet
Q: I encounter segment fault
at the beginning.
A: A compatibility issue has been identified between MXNet and opencv-python 3.0+. We suggest that you always import cv2
first before import mxnet
in the entry script.
Q: I find the training speed becomes slower when training for a long time.
A: It has been identified that MXNet on Windows has this problem. So we recommend to run this program on Linux. You could also stop it and resume the training process to regain the training speed if you encounter this problem.
Q: Can you share your caffe implementation?
A: Due to several reasons (code is based on a old, internal Caffe, port to public Caffe needs extra work, time limit, etc.). We do not plan to release our Caffe code. Since current MXNet convolution implementation is very similar to Caffe (almost the same), it is easy to port to Caffe by yourself, the core CUDA code could be kept unchanged. Anyone who wish to do it is welcome to make a pull request.