Download and extractaugmented segmentation data(Thanks to DrSleep), specifying the location with --data_dir and --label_data_dir(namely, $data_dir/$label_data_dir).
For inference the trained model with 76.42% mIoU on the Pascal VOC 2012 validation dataset
is availablehere. Download and extract to--model_dir.
Here, --pre_trained_model contains the pre-trained Resnet model, whereas--model_dir contains the trained DeepLabv3 checkpoints.
If --model_dir contains the valid checkpoints, the model is trained from the
specified checkpoint in --model_dir.
You can see other options with the following command:
python train.py --help
The training process can be visualized with Tensor Board as follow:
tensorboard --logdir MODEL_DIR
Evaluation
To evaluate how model perform, one can use the following command:
python evaluate.py --help
The current best model build by this implementation achieves 76.42% mIoU on the Pascal VOC 2012
validation dataset.
Method
OS
mIOU
paper
MG(1,2,4)+ASPP(6,12,18)+Image Pooling
16
77.21%
repo
MG(1,2,4)+ASPP(6,12,18)+Image Pooling
16
76.42%
Here, the above model was trained about 9.5 hours (with Tesla V100 and r1.6) with following parameters: