An implementation of DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs
You can download pretrianed model on COCO here.
python train.py
- Performance in mean IoU on VOC 12 val set (iteration: 20k, single-scale input for training)
| input | Deeplab_Pytorch |
|---|---|
| x1 | 72.05% |
| x(1, 3/4, 1/2) | 73.00% |
note: using batchsize=12for training,
you can download the model here
This code is heavily borrowed from Pytorch-Deeplab.