Training:
- Tensorflow (tested with version 0.12)
Evaluation:
- Matlab
- MatConvNet
- Prepare Cityscapes dataset: Convert background label
255to19 - Download models and place them inside the 'spatialAnticipationNetwork'-root folder.
- Adapt the paths in
train.py - Train the model using
python train.py - Adapt the paths in
eval.py - Predict labels on the validation set
python eval.py - Compute IoU and F1-scores by using
./matlab/evaluateAllResults.mafter adapting paths
The tensorflow code in this repository was written by modifying a duplicate of DrSleep's-deeplab-tensorflow project. The Matlab evaluation scripts were written by modifying Liang-Chieh Chen's deeplab-public-ver2