Exploring Local Detail Perception for Scene Sketch Semantic Segmentation

Code release for "Exploring Local Detail Perception for Scene Sketch Semantic Segmentation" (IEEE TIP)

Requirements

• Create a conda environment from the environment.yml file:

conda env create -f environment.yml

• Activate the environment:

conda activate LDP

Preparations

• Get the code:

git clone https://github.com/drcege/Local-Detail-Perception && cd Local-Detail-Perception

• Download datasets from releases and place them under the datasets directory following its instructions.

• Generate ImageNet pre-trained "ResNet-101" model in TensorFlow version for initial training and place it under the resnet_pretrained_model directory. This can be obtained following the instructions in chenxi116/TF-resnet. For convenience, the converted model can be downloaded from here.

Training

python3 segment_main.py --mode=train --run_name=LDP 

Evaluation

python3 segment_main.py --mode=test --run_name=LDP

Credits

• The ResNet-101 model pre-trained on ImageNet in TensorFlow is created by chenxi116
• The code for the DeepLab model is authored by Tensorflow authors and chenxi116
• The repository is developed based on SketchyScene