Benchmark and Model Zoo

Common settings

  • We use distributed training with 4 GPUs by default.

  • All pytorch-style pretrained backbones on ImageNet are train by ourselves, with the same procedure in the paper. Our ResNet style backbone are based on ResNetV1c variant, where the 7x7 conv in the input stem is replaced with three 3x3 convs.

  • For the consistency across different hardwares, we report the GPU memory as the maximum value of torch.cuda.max_memory_allocated() for all 4 GPUs with torch.backends.cudnn.benchmark=False. Note that this value is usually less than what nvidia-smi shows.

  • We report the inference time as the total time of network forwarding and post-processing, excluding the data loading time. Results are obtained with the script tools/ which computes the average time on 200 images with torch.backends.cudnn.benchmark=False.

  • There are two inference modes in this framework.

    • slide mode: The test_cfg will be like dict(mode='slide', crop_size=(769, 769), stride=(513, 513)).

      In this mode, multiple patches will be cropped from input image, passed into network individually. The crop size and stride between patches are specified by crop_size and stride. The overlapping area will be merged by average

    • whole mode: The test_cfg will be like dict(mode='whole').

      In this mode, the whole imaged will be passed into network directly.

      By default, we use slide inference for 769x769 trained model, whole inference for the rest.

  • For input size of 8x+1 (e.g. 769), align_corner=True is adopted as a traditional practice. Otherwise, for input size of 8x (e.g. 512, 1024), align_corner=False is adopted.



Please refer to FCN for details.


Please refer to PSPNet for details.


Please refer to DeepLabV3 for details.


Please refer to PSANet for details.


Please refer to DeepLabV3+ for details.


Please refer to UPerNet for details.

NonLocal Net

Please refer to NonLocal Net for details.


Please refer to EncNet for details.


Please refer to CCNet for details.


Please refer to DANet for details.


Please refer to HRNet for details.


Please refer to GCNet for details.


Please refer to ANN for details.


Please refer to OCRNet for details.


Please refer to Fast-SCNN for details.


Please refer to ResNeSt for details.

Semantic FPN

Please refer to Semantic FPN for details.


Please refer to PointRend for details.


Please refer to MobileNetV2 for details.


Please refer to EMANet for details.


Please refer to DNLNet for details.


Please refer to CGNet for details.

Mixed Precision (FP16) Training

Please refer Mixed Precision (FP16) Training for details.

Speed benchmark


  • 8 NVIDIA Tesla V100 (32G) GPUs
  • Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz

Software environment

  • Python 3.7
  • PyTorch 1.5
  • CUDA 10.1
  • CUDNN 7.6.03
  • NCCL 2.4.08

Training speed

For fair comparison, we benchmark all implementations with ResNet-101V1c. The input size is fixed to 1024x512 with batch size 2.

The training speed is reported as followed, in terms of second per iter (s/iter). The lower, the better.

Implementation PSPNet (s/iter) DeepLabV3+ (s/iter)
MMSegmentation 0.83 0.85
SegmenTron 0.84 0.85
CASILVision 1.15 N/A
vedaseg 0.95 1.25

Note: The output stride of DeepLabV3+ is 8.