We provide training resources for participants on the OpenI platform where the participants can use two v100 to train their models. If you want to use the GPUs we provided, please register your account in https://openi.org.cn/ and submit your username and mails to here. You can check if you have successfully enabled the permission here.
1. migrate the code from github
You can migrate the United-Perception repo to openi as follows and we recommend that you switch to the AAAI_challenge branch, which is specially prepared for the OpenI platform.
2. build your dataset
You can build your own datasets by associating datasets we provided.
3. Debug and Train
3.1 Debug
The OpenI provides convenient debugging and training functions. Participants can debug their code by following the steps below.
You can choose your most similar tools to debug your code.
3.2 Train
During training, your code is saved in /tmp/code/ by default, datasets are saved in /tmp/dataset/ by default, and the outputs is saved in /tmp/output/ by default. More details can be seen here. You can follow these steps to train your model.
The training Image is：dockerhub.pcl.ac.cn:5000/user-images/openi:1.3.0
You can view the training results and download the model files here.
