DeepLabV3

Last updated on Feb 19, 2021

DeepLabV3 (R101-DC5)

Parameters
Backbone Layers 101
Training Data Cityscapes
Training Resources 8 NVIDIA V100 GPUs
Training Time

Architecture 1x1 Convolution, ASPP, Batch Normalization, Dilated Convolution, ResNet
Max Iter 90000
ASPP Dropout 0.1
Backbone Layers 101
Output Resolution 1024×2048
SHOW MORE
SHOW LESS
DeepLabV3 (R103-DC5)

Parameters 58 Million
Backbone Layers 101
File Size 222.37 MB
Training Data Cityscapes
Training Resources 8 NVIDIA V100 GPUs
Training Time

Architecture 1x1 Convolution, ASPP, Batch Normalization, Dilated Convolution
ID 28041665
Max Iter 90000
ASPP Dropout 0.1
Backbone Layers 101
Output Resolution 1024×2048
SHOW MORE
SHOW LESS
README.md

Summary

DeepLabv3 is a semantic segmentation architecture that improves upon DeepLabv2 with several modifications.To handle the problem of segmenting objects at multiple scales, modules are designed which employ atrous convolution in cascade or in parallel to capture multi-scale context by adopting multiple atrous rates. Furthermore, the Atrous Spatial Pyramid Pooling module from DeepLabv2 augmented with image-level features encoding global context and further boost performance.

How do I evaluate this model?

Model evaluation can be done as follows:

cd /path/to/detectron2/projects/DeepLab
python train_net.py --config-file configs/Cityscapes-SemanticSegmentation/deeplab_v3_plus_R_103_os16_mg124_poly_90k_bs16.yaml --eval-only MODEL.WEIGHTS /path/to/model_checkpoint

How do I train this model?

To train a model with 8 GPUs run:

cd /path/to/detectron2/projects/DeepLab
python train_net.py --config-file configs/Cityscapes-SemanticSegmentation/deeplab_v3_plus_R_103_os16_mg124_poly_90k_bs16.yaml --num-gpus 8

You can follow the Getting Started guide on Colab to see how to train a model.

You can also read the official Detectron2 documentation.

Citation

@article{deeplabv32018,
  title={Rethinking atrous convolution for semantic image segmentation},
  author={Chen, Liang-Chieh and Papandreou, George and Schroff, Florian and Adam, Hartwig},
  journal={arXiv:1706.05587},
  year={2017}
}

Results

Semantic Segmentation on Cityscapes val

Semantic Segmentation
BENCHMARK MODEL METRIC NAME METRIC VALUE GLOBAL RANK
Cityscapes val DeepLabV3 (R103-DC5) mIoU 78.5 # 3
Cityscapes val DeepLabV3 (R101-DC5) mIoU 76.7 # 5