Notice(2019.11.2)

This repo was built back two years ago when there were no pytorch detection implementation that can achieve reasonable performance. At this time, there are many better repos out there, for example:

• detectron2(It's a masterpiece.)
• mmdetection

Therefore, this repo will not be actively maintained.

Important notice:

If you used the master branch before Sep. 26 2017 and its corresponding pretrained model, PLEASE PAY ATTENTION: The old master branch in now under old_master, you can still run the code and download the pretrained model, but the pretrained model for that old master is not compatible to the current master!

The main differences between new and old master branch are in this two commits: 9d4c24e, c899ce7 The change is related to this issue; master now matches all the details in tf-faster-rcnn so that we can now convert pretrained tf model to pytorch model.

pytorch-faster-rcnn

A pytorch implementation of faster RCNN detection framework based on Xinlei Chen's tf-faster-rcnn. Xinlei Chen's repository is based on the python Caffe implementation of faster RCNN available here.

Note: Several minor modifications are made when reimplementing the framework, which give potential improvements. For details about the modifications and ablative analysis, please refer to the technical report An Implementation of Faster RCNN with Study for Region Sampling. If you are seeking to reproduce the results in the original paper, please use the official code or maybe the semi-official code. For details about the faster RCNN architecture please refer to the paper Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks.

Detection Performance

The current code supports VGG16, Resnet V1 and Mobilenet V1 models. We mainly tested it on plain VGG16 and Resnet101 architecture. As the baseline, we report numbers using a single model on a single convolution layer, so no multi-scale, no multi-stage bounding box regression, no skip-connection, no extra input is used. The only data augmentation technique is left-right flipping during training following the original Faster RCNN. All models are released.

With VGG16 (conv5_3):

• Train on VOC 2007 trainval and test on VOC 2007 test, 71.22(from scratch) 70.75(converted) (70.8 for tf-faster-rcnn).
• Train on VOC 2007+2012 trainval and test on VOC 2007 test (R-FCN schedule), 75.33(from scratch) 75.27(converted) (75.7 for tf-faster-rcnn).
• Train on COCO 2014 trainval35k and test on minival (900k/1190k) 29.2(from scratch) 30.1(converted) (30.2 for tf-faster-rcnn).

With Resnet101 (last conv4):

• Train on VOC 2007 trainval and test on VOC 2007 test, 75.29(from scratch) 75.76(converted) (75.7 for tf-faster-rcnn).
• Train on VOC 2007+2012 trainval and test on VOC 2007 test (R-FCN schedule), 79.26(from scratch) 79.78(converted) (79.8 for tf-faster-rcnn).
• Train on COCO 2014 trainval35k and test on minival (800k/1190k), 35.1(from scratch) 35.4(converted) （35.4 for tf-faster-rcnn).

More Results:

• Train Mobilenet (1.0, 224) on COCO 2014 trainval35k and test on minival (900k/1190k), 21.4(from scratch), 21.9(converted) （21.8 for tf-faster-rcnn).
• Train Resnet50 on COCO 2014 trainval35k and test on minival (900k/1190k), 32.4(converted) (32.4 for tf-faster-rcnn).
• Train Resnet152 on COCO 2014 trainval35k and test on minival (900k/1190k), 36.7(converted) (36.1 for tf-faster-rcnn).

Approximate baseline setup from FPN (this repository does not contain training code for FPN yet):

• Train Resnet50 on COCO 2014 trainval35k and test on minival (900k/1190k), ~~34.2~~.
• Train Resnet101 on COCO 2014 trainval35k and test on minival (900k/1190k), ~~37.4~~.
• Train Resnet152 on COCO 2014 trainval35k and test on minival (900k/1190k), ~~38.2~~.

Note:

• Due to the randomness in GPU training especially for VOC, the best numbers are reported (with 2-3 attempts) here. According to Xinlei's experience, for COCO you can almost always get a very close number (within ~0.2%) despite the randomness.

• The numbers are obtained with the default testing scheme which selects region proposals using non-maximal suppression (TEST.MODE nms), the alternative testing scheme (TEST.MODE top) will likely result in slightly better performance (see report, for COCO it boosts 0.X AP).

• Since we keep the small proposals (< 16 pixels width/height), our performance is especially good for small objects.

• We do not set a threshold (instead of 0.05) for a detection to be included in the final result, which increases recall.

• Weight decay is set to 1e-4.

• For other minor modifications, please check the report. Notable ones include using crop_and_resize, and excluding ground truth boxes in RoIs during training.

• For COCO, we find the performance improving with more iterations, and potentially better performance can be achieved with even more iterations.

• For Resnets, we fix the first block (total 4) when fine-tuning the network, and only use crop_and_resize to resize the RoIs (7x7) without max-pool (which Xinlei finds useless especially for COCO). The final feature maps are average-pooled for classification and regression. All batch normalization parameters are fixed. Learning rate for biases is not doubled.

• For Mobilenets, we fix the first five layers when fine-tuning the network. All batch normalization parameters are fixed. Weight decay for Mobilenet layers is set to 4e-5.

• For approximate FPN baseline setup we simply resize the image with 800 pixels, add 32^2 anchors, and take 1000 proposals during testing.

• Check out here/here/here for the latest models, including longer COCO VGG16 models and Resnet ones.

| :-------------------------:|:-------------------------: Displayed Ground Truth on Tensorboard | Displayed Predictions on Tensorboard

Additional features

Additional features not mentioned in the report are added to make research life easier:

• Support for train-and-validation. During training, the validation data will also be tested from time to time to monitor the process and check potential overfitting. Ideally training and validation should be separate, where the model is loaded every time to test on validation. However Xinlei have implemented it in a joint way to save time and GPU memory. Though in the default setup the testing data is used for validation, no special attempts is made to overfit on testing set.
• Support for resuming training. Xinlei tried to store as much information as possible when snapshoting, with the purpose to resume training from the latest snapshot properly. The meta information includes current image index, permutation of images, and random state of numpy. However, when you resume training the random seed for tensorflow will be reset (not sure how to save the random state of tensorflow now), so it will result in a difference. Note that, the current implementation still cannot force the model to behave deterministically even with the random seeds set. Suggestion/solution is welcome and much appreciated.
• Support for visualization. The current implementation will summarize ground truth boxes, statistics of losses, activations and variables during training, and dump it to a separate folder for tensorboard visualization. The computing graph is also saved for debugging.

Prerequisites

• A basic pytorch installation. The code follows 1.0. If you are using old 0.1.12 or 0.2 or 0.3 or 0.4, you can checkout the corresponding branch.
• Torchvision 0.3. This code uses torchvision.ops for nms, roi_pool and roi_align
• Python packages you might not have: opencv-python, easydict (similar to py-faster-rcnn). For easydict make sure you have the right version. Xinlei uses 1.6.
• tensorboard-pytorch to visualize the training and validation curve. Please build from source to use the latest tensorflow-tensorboard.
• ~~Docker users: Since the recent upgrade, the docker image on docker hub (https://hub.docker.com/r/mbuckler/tf-faster-rcnn-deps/) is no longer valid. However, you can still build your own image by using dockerfile located at docker folder (cuda 8 version, as it is required by Tensorflow r1.0.) And make sure following Tensorflow installation to install and use nvidia-docker[https://github.com/NVIDIA/nvidia-docker]. Last, after la