HaozheLiu-ST
commented
1 year ago Hello, you can directly run the code in NICOChallenge-OOD-Classification/moco/ . For the model weights, since this challenge asks us to submit the code rather than model weights to ensure fairness, we just archive our code without the training weights. Due to the previous storage re-schedule, we cannot provide the trained weight. If you have any questions on running NICOChallenge-OOD-Classification/moco/ , please let me know. I think it is a good idea to employ self-supervised learning on NICO++. Compared with the random initialization, self-supervised learning can improve the result significantly.
JoyHuYY1412
Hi Haozhe, thank you for your codes! As I read from your paper and the ReadMe, you used MOCO to first train the NOCO++ dataset in the self-supervised way. Am I right? Could you please share your pre-trained Self-supervised models based on MOCO? Or did you find some meaningful clusters after self-supervised pre-training? I am also interested in performing self-supervised learning on NICO++, too. However, I am afraid that the dataset is not large enough. It would be very nice if you could provide some trained models or give me some advice.
I really appreciate your reply!