The problem of retraining

[melon-coder]

I want to train a different type of action from the ava data set, how should I train it? Is it enough to prepare the corresponding comments directly in the data preparation part? What else needs to be changed?

[melon-coder]

Is it okay not to prepare ava_val_det_person_bbox.json, ava_train_det_object_bbox.json, and ava_val_det_object_bbox.json during training? Because the csv and converted json files already contain action information?

[melon-coder]

在训练期间不准备ava_val_det_person_bbox.json, ava_train_det_object_bbox.json, 和 可以ava_val_det_object_bbox.json吗？因为 csv 和转换后的 json 文件已经包含动作信息？

Can you tell me how to generate these json files? I hope to hear from you, thank you very much!

[shnulailai]

Ask for the same question.