The objective of this repository is to replicate and extend upon Cross-Domain Weakly-Supervised Object Detection through Progressive Domain Adaptation). The implementation is based upon High quality, fast, modular reference implementation of SSD in PyTorch and Unofficial pytorch implementation of AdaIN. The results of this project can be found in our academic report paper here.
Detection results before (baseline) and after (adain).
For a thorough explanation on how we executed, or how to replicate our experiment please refer to these notebooks: ssd+adain, cyclegan. For further details on how to setup your environment you may check the original authors recommendations.
We make available the models as explained in the paper linked above. The performance is evaluated on Clipart1k test split, you can find more about it here
| Name | Configuration | mAP | Download |
|---|---|---|---|
| baseline | setup_3 | 25.86 | model |
| cyclegan | eval_1 | 31.95 | model |
| adain | style_1 | 39.25 | model |