Mingxing comments

Results 20 comments of


                                            Mingxing

How to hyper parameters tuning for small object (objects < 32x32)

Do you mean all your objects are about 32x32, or do you mean these objects can have different size? Also, what's your image size?

Inference Differences

Hi @cgeller , could you paste your complete command line for these two approaches? One potential issues is that when you export the saved model, it already uses the max_instance...

How can I train det-AdvProp in my own dataset ？

Hi @secortot , you are right, advprop is not open sourced: it is quite expensive to run. I will ask my collaborators to try to open source it when he...

KeyError: 'trainable_weights'

Could you paste the full command line? I don't understand your question.

what is the actual meaning of num_classes ?

Sorry, it should be 20, and 0 is always added for background.

what is the actual meaning of num_classes ?

Actually, it should be 21 (the length of the label map, including background). Sorry, I gave a wrong information.

Gradual training strategy

Yeah, sounds like a good (and simple optimization) to speed up the training, but I am not sure about the quality impact.

State of the art paper which gives 2+ map boost on coco with just augmentation

Hi @kartik4949, it would be great if you could help add that :)

State of the art paper which gives 2+ map boost on coco with just augmentation

Wow, this is great. could you check in this augmentation? I can't wait to re-train the models with this new augmentation.

Unable to do evaluation on the trained model.

Hi @Koteswara-ML Apparently, it is because num_classes is different at train and eval time: one is 90 classes and the other is 1 classes.