I have two questions. Can we use pre-trained models such as MOCO_BO, MOCO_Imagenet or others in featues files? Can we use unlabeled set of 11 buckers(300G for each bucket) for unsupervised training? Looking forward to your reply.
- No. It’s emphasized in README of your starter kit, we don’t allow pre-trained models. These pre-trained models are provided as baselines for you as a reference. However, to qualify Stage 2, you have to train your models from scratch.
- In terms of this challenge, the answer is no. However, you are more than welcome to try on these unlabeled data for unsupervised learning problems and feel free to make use of them on your own research!
Strongly recommend you to join our discord or slack channel for more prompt reply from us. The invitation links can be found at the bottom of the challenge overview page.