gaopengcuhk / Pretrained-Pix2Seq

Replication of Pix2Seq with Pretrained Model

Pretrained-Pix2Seq

We provide the pre-trained model of Pix2Seq. This version contains new data augmentation. The model is trained for 300 epochs and can acheive 37 mAP without beam search or neucles search.

Installation

Install PyTorch 1.5+ and torchvision 0.6+ (recommend torch1.8.1 torchvision 0.8.0)

Install pycocotools (for evaluation on COCO):

pip install -U 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'

That's it, should be good to train and evaluate detection models.

Data preparation

Download and extract COCO 2017 train and val images with annotations from http://cocodataset.org. We expect the directory structure to be the following:

path/to/coco/
  annotations/  # annotation json files
  train2017/    # train images
  val2017/      # val images

Training

First link coco dataset to the project folder

ln -s /path/to/coco ./coco 

Training

sh train.sh --model pix2seq --output_dir /path/to/save

Evaluation

sh train.sh --model pix2seq --output_dir /path/to/save --resume /path/to/checkpoints --eval

COCO

Method	backbone	Epoch	Batch Size	AP	AP50	AP75	Weights
Pix2Seq	R50	300	32	37.0	53.4	39.4	weight

Contributor

Qiu Han, Peng Gao, Jingqiu Zhou(Beam Search)

Acknowledegement

Pix2Seq, DETR