Efficient Mirror Detection via Multi-level Heterogeneous Learning (AAAI23, ORAL)

Authors: Ruozhen He, Jiaying Lin, and Rynson Lau [Paper Link]

Dataset

• Download the PMD dataset at here and MSD dataset at here.

Experimental Results

• We provide HetNet training weights on MSD and PMD presented in the papers.
• We provide HetNet testing maps on MSD and PMD presented in the papers.

Code

Pretrained weights

• The pretrained weight can be found here: ResNeXt-101.
• Put the downloaded weights under ./resnext/.

Train

• Download the dataset and pretrained model.
• Modify the path in train.py.
• Run python train.py.

Test and Evaluate

• Modify the path and filename.
• Run python test.py.
• Run python eval.py.
• The efficiency evaluation is supported by the ptflops toolkit. To evaluate the FLOPs and Parameters, run python eval_complexity.py.