This is an unofficial pytorch implementation of ATSS(retina) object detection as described in Bridging the Gap Between Anchor-based and Anchor-free Detection via Adaptive Training Sample Selection by Shifeng Zhang, Cheng Chi, Yongqiang Yao, Zhen Lei, Stan Z. Li.

tqdm
pyyaml
numpy
opencv-python
pycocotools
torch >= 1.5
torchvision >=0.6.0

we trained this repo on 4 GPUs with batch size 32(8 image per node).the total epoch is 24(about 180k iter),Adam with cosine lr decay is used for optimizing. finally, this repo achieves 39.6 mAp at 640px(max side) resolution with resnet50 backbone.(about 42.95fps)

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.396
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.589
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.426
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.218
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.434
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.546
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.322
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.513
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.557
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.354
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.611
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.723

the main difference is about the input resolution.the original implement use min_thresh and max_thresh to keep the short side of the input image larger than min_thresh while keep the long side smaller than max_thresh.for simplicity we fix the long side a certain size, then we resize the input image while keep the width/height ratio, next we pad the short side.the final width and height of the input are same.

for now we only support coco detection data.

• modify main.py (modify config file path)

from solver.ddp_mix_solver import DDPMixSolver
if __name__ == '__main__':
    processor = DDPMixSolver(cfg_path="your own config path") 
    processor.run()

• custom some parameters in config.yaml

model_name: atss_retina
data:
  train_annotation_path: data/annotations/instances_train2017.json
#  train_annotation_path: data/annotations/instances_val2017.json
  val_annotation_path: data/annotations/instances_val2017.json
  train_img_root: data/train2017
#  train_img_root: data/val2017
  val_img_root: data/val2017
  max_thresh: 640
  use_crowd: False
  batch_size: 8
  num_workers: 4
  debug: False
  remove_blank: Ture

model:
  num_cls: 80
  anchor_sizes: [32, 64, 128, 256, 512]
  strides: [8, 16, 32, 64, 128]
  backbone: resnet50
  pretrained: True
  top_k: 9
  alpha: 0.25
  gamma: 2.0
  iou_type: giou
  iou_loss_weight: 0.5
  reg_loss_weight: 1.15
  iou_loss_type: centerness
  conf_thresh: 0.05
  nms_iou_thresh: 0.5
  max_det: 300

optim:
  optimizer: Adam
  lr: 0.0001
  milestones: [18,24]
  warm_up_epoch: 0
  weight_decay: 0.0001
  epochs: 24
  sync_bn: True
  amp: True
val:
  interval: 1
  weight_path: weights


gpus: 0,1,2,3

• run train scripts

nohup python -m torch.distributed.launch --nproc_per_node=4 main.py >>train.log 2>&1 &

• Color Jitter
• Perspective Transform
• Mosaic Augment
• MixUp Augment
• IOU GIOU DIOU CIOU
• Warming UP
• Cosine Lr Decay
• EMA(Exponential Moving Average)
• Mixed Precision Training (supported by apex)
• Sync Batch Normalize
• PANet(neck)
• BiFPN(EfficientDet neck)
• VOC data train\test scripts
• custom data train\test scripts
• MobileNet Backbone support

many helps from this work,Edwardwaw/atss_retinanet (mAP 39.4)