This project is our implementation of "Sketch Transformer: Asymmetrical Disentanglement Learning from Dynamic Synthesis" (ACM MM2022) and the extension of "SketchTrans: Disentangled Prototype Learning With Transformer for Sketch-Photo Recognition" (TPAMI2024).

1. Prepare Datasets

(1) Category-level datasets: link

(2)Instance-level Datasets: PKU-Sketch dataset: https://www.pkuml.org/resources/pkusketchreid-dataset.html; QMUL: https://sketchx.eecs.qmul.ac.uk/downloads/

2. Running Train and Test

(1) Train and Test for category-level sketch-photo recognition: sh ./category_dist_train.sh;

(2)Train and Test for instance-level sketch-photo recognition: sh ./instance_dist_train.sh.

3. Trained Models

Our trained models and the pretrained weights of the generator G can be downloaded from baidu netdisk (the extraction code is d28h).

4. Citation

@ARTICLE{chen_sketchtrans_tpami, author={Chen, Cuiqun and Ye, Mang and Qi, Meibin and Du, Bo}, journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, title={SketchTrans: Disentangled Prototype Learning With Transformer for Sketch-Photo Recognition}, year={2024}, volume={46}, number={5}, pages={2950-2964} }

@inproceedings{chen2022sketch, title={Sketch transformer: Asymmetrical disentanglement learning from dynamic synthesis}, author={Chen, Cuiqun and Ye, Mang and Qi, Meibin and Du, Bo}, booktitle={Proceedings of the 30th ACM International Conference on Multimedia}, pages={4012--4020}, year={2022} }

5. License

The code is distributed under the MIT License. See LICENSE for more information.