Our team is very interested in your project and also has some questions about replicating the code.
In the project description, running the project is very simple, just Run run.sh . However, you have pointed out that the checkpoint --checkpoint "output/pretrained/checkpoint_32.pth" for evaluation is missing.
Upon reading the introduction of the experimental training in the paper, you found that the training is divided into two stages. In the first stage, the entire model is trained, and in the second stage, AUL is trained. We are very interested in this staged training strategy. Unfortunately, we did not find any details about the training strategies for these two phases in this project, nor did we find the training objectives for each phase.
What training objectives are included in the first training phase, and what are the training objectives for the first phase of training?
Are the weights initialized randomly or from pre-trained weights of VLP?
We think this project is very valuable, and we are eagerly looking forward to receiving the answers to these questions.
lizimo-n
commented
5 months ago
Our team is very interested in your project and also has some questions about replicating the code.
In the project description, running the project is very simple, just
Run run.sh. However, you have pointed out that the checkpoint--checkpoint "output/pretrained/checkpoint_32.pth"for evaluation is missing.Upon reading the introduction of the experimental training in the paper, you found that the training is divided into two stages. In the first stage, the entire model is trained, and in the second stage, AUL is trained. We are very interested in this staged training strategy. Unfortunately, we did not find any details about the training strategies for these two phases in this project, nor did we find the training objectives for each phase.
What training objectives are included in the first training phase, and what are the training objectives for the first phase of training?
Are the weights initialized randomly or from pre-trained weights of VLP?
We think this project is very valuable, and we are eagerly looking forward to receiving the answers to these questions.