Retinanet Pretrained models

[Darshan2701]

❓ Questions and Help

Hi,

I wanted to train retinanet 50 FPN with 2 classes. I downloaded the pre-trained models from #102 and chopped off cls_logits variables to fit my classes. Started train_net.py and it did not train at all, rather it directly went ahead and did the testing part on my test dataset.
Could you please let me know if I am missing some procedure.

[Darshan2701]

I think i need to delete the optimizer and scheduler dict, if I am not wrong before using the pre-trained model.

[Darshan2701]

Update: This did work on my dataset but the inference speed is slower than the faster rcnn resnet50 FPN models!
Any pointers as to why this is happening?

[tangze5258]

I think i need to delete the optimizer and scheduler dict, if I am not wrong before using the pre-trained model.

@Darshan2701 hi，i have same prolerms like you. i want to retrain my new datasets based on pretrained model_final.pth.And i delete optimizer， scheduler and iteration dict too, it did work when run train_net.py, but finally, has no new model_final.pth file generated,what's wrong ? thx....

[Darshan2701]

@tangze5258 Were you able to solve this?
Sorry for the delayed ask but.

[Darshan2701]

@tangze5258 I think u need to save the new model_final.pth in a different folder than the one used already as a new pre-trained model.