Open Andreahernandezt opened 2 years ago
@Andreahernandezt It would be great if you can share the procedure for training on your custom dataset. Apart from editing the configuration files, what other are important factors?
If you have posted your work somewhere else, a link would be valuable here.
Looking forward to hear from you
I am using my own dataset and reading the repository issues I saw people saying that they got better results without loading the model coco_dla_1x or 2x I did indeed train with and without fine-tunning, and the results were quite different. When training without fine-tunning achieved with my database a performance quite similar to the paper. However, when training with fine-tunning I got bad results.
Is there any reason why this happens?