Awesome

An Empirical Study of Self-supervised Learning Approaches for Object Detection with Transformers

Self-supervised learning (SSL) methods such as masked language modeling have shown massive performance gains by pretraining transformer models for a variety of natural language processing tasks. The follow-up research adapted similar methods like masked image modeling in vision transformer and demonstrated improvements in the image classification task. Such simple self-supervised methods are not exhaustively studied for object detection transformers (DETR, Deformable DETR) as their transformer encoder modules take input in the convolutional neural network (CNN) extracted feature space rather than the image space as in general vision transformers. However, the CNN feature maps still maintain the spatial relationship and we utilize this property to design self-supervised learning approaches to train the encoder of object detection transformers in pretraining and multi-task learning settings. We explore common self-supervised methods based on image reconstruction, masked image modeling and jigsaw. Preliminary experiments in the iSAID dataset demonstrate faster convergence of DETR in the initial epochs in both pretraining and multi-task learning settings; nonetheless, similar improvement is not observed in the case of multi-task learning with Deformable DETR. The code for our experiments with DETR and Deformable DETR are available at https://github.com/gokulkarthik/detr and https://github.com/gokulkarthik/Deformable-DETR respectively.

ArXiv: https://arxiv.org/abs/2205.05543

Authors: Gokul Karthik Kumar, Sahal Shaji Mullappilly, Abhishek Singh Gehlot

Work Adapted from Deformable DETR:

By Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, Jifeng Dai.

This repository is an official implementation of the paper Deformable DETR: Deformable Transformers for End-to-End Object Detection.

Introduction

TL; DR. Deformable DETR is an efficient and fast-converging end-to-end object detector. It mitigates the high complexity and slow convergence issues of DETR via a novel sampling-based efficient attention mechanism.

Abstract. DETR has been recently proposed to eliminate the need for many hand-designed components in object detection while demonstrating good performance. However, it suffers from slow convergence and limited feature spatial resolution, due to the limitation of Transformer attention modules in processing image feature maps. To mitigate these issues, we proposed Deformable DETR, whose attention modules only attend to a small set of key sampling points around a reference. Deformable DETR can achieve better performance than DETR (especially on small objects) with 10× less training epochs. Extensive experiments on the COCO benchmark demonstrate the effectiveness of our approach.

License

This project is released under the Apache 2.0 license.

Changelog

See changelog.md for detailed logs of major changes.

Citing Deformable DETR

If you find Deformable DETR useful in your research, please consider citing:

@article{zhu2020deformable,
  title={Deformable DETR: Deformable Transformers for End-to-End Object Detection},
  author={Zhu, Xizhou and Su, Weijie and Lu, Lewei and Li, Bin and Wang, Xiaogang and Dai, Jifeng},
  journal={arXiv preprint arXiv:2010.04159},
  year={2020}
}

Main Results

Note:

1. All models of Deformable DETR are trained with total batch size of 32.
2. Training and inference speed are measured on NVIDIA Tesla V100 GPU.
3. "Deformable DETR (single scale)" means only using res5 feature map (of stride 32) as input feature maps for Deformable Transformer Encoder.
4. "DC5" means removing the stride in C5 stage of ResNet and add a dilation of 2 instead.
5. "DETR-DC5+" indicates DETR-DC5 with some modifications, including using Focal Loss for bounding box classification and increasing number of object queries to 300.
6. "Batch Infer Speed" refer to inference with batch size = 4 to maximize GPU utilization.
7. The original implementation is based on our internal codebase. There are slight differences in the final accuracy and running time due to the plenty details in platform switch.

Installation

Requirements

• Linux, CUDA>=9.2, GCC>=5.4

• Python>=3.7

  We recommend you to use Anaconda to create a conda environment:

  conda create -n deformable_detr python=3.7 pip
  

  Then, activate the environment:

  conda activate deformable_detr
  

• PyTorch>=1.5.1, torchvision>=0.6.1 (following instructions here)

  For example, if your CUDA version is 9.2, you could install pytorch and torchvision as following:

  conda install pytorch=1.5.1 torchvision=0.6.1 cudatoolkit=9.2 -c pytorch
  

• Other requirements

  pip install -r requirements.txt
  

Compiling CUDA operators

cd ./models/ops
sh ./make.sh
# unit test (should see all checking is True)
python test.py

Usage

Dataset preparation

Please download COCO 2017 dataset and organize them as following:

code_root/
└── data/
    └── coco/
        ├── train2017/
        ├── val2017/
        └── annotations/
        	├── instances_train2017.json
        	└── instances_val2017.json

Training

Training on single node

For example, the command for training Deformable DETR on 8 GPUs is as following:

GPUS_PER_NODE=8 ./tools/run_dist_launch.sh 8 ./configs/r50_deformable_detr.sh

Training on multiple nodes

For example, the command for training Deformable DETR on 2 nodes of each with 8 GPUs is as following:

On node 1:

MASTER_ADDR=<IP address of node 1> NODE_RANK=0 GPUS_PER_NODE=8 ./tools/run_dist_launch.sh 16 ./configs/r50_deformable_detr.sh

On node 2:

MASTER_ADDR=<IP address of node 1> NODE_RANK=1 GPUS_PER_NODE=8 ./tools/run_dist_launch.sh 16 ./configs/r50_deformable_detr.sh

Training on slurm cluster

If you are using slurm cluster, you can simply run the following command to train on 1 node with 8 GPUs:

GPUS_PER_NODE=8 ./tools/run_dist_slurm.sh <partition> deformable_detr 8 configs/r50_deformable_detr.sh

Or 2 nodes of each with 8 GPUs:

GPUS_PER_NODE=8 ./tools/run_dist_slurm.sh <partition> deformable_detr 16 configs/r50_deformable_detr.sh

Some tips to speed-up training

• If your file system is slow to read images, you may consider enabling '--cache_mode' option to load whole dataset into memory at the beginning of training.
• You may increase the batch size to maximize the GPU utilization, according to GPU memory of yours, e.g., set '--batch_size 3' or '--batch_size 4'.

Evaluation

You can get the config file and pretrained model of Deformable DETR (the link is in "Main Results" session), then run following command to evaluate it on COCO 2017 validation set:

<path to config file> --resume <path to pre-trained model> --eval

You can also run distributed evaluation by using ./tools/run_dist_launch.sh or ./tools/run_dist_slurm.sh.