YanxingLiu / SAE-FSDet

Code for reproducing the results in our TGRS-2024 paper Few-Shot Object Detection in Remote Sensing Images via Label-Consistent Classifier and Gradual Regression.
Apache License 2.0
19 stars 1 forks source link

How to Filter and Save Results for a Specific Class in Few-Shot Detection #13

Open MarineMehrabyan opened 1 month ago

MarineMehrabyan commented 1 month ago

I want to detect only one class. For example, if I trained the base model on the 'dog' class and then fine-tuned it using a few-shot approach for the 'cat' class, how can I save only the images where the model detects the 'cat' class? Thank you for your time.

YanxingLiu commented 1 month ago

I want to detect only one class. For example, if I trained the base model on the 'dog' class and then fine-tuned it using a few-shot approach for the 'cat' class, how can I save only the images where the model detects the 'cat' class? Thank you for your time.

You can modify the detection head to detect novel classes

MarineMehrabyan commented 1 month ago

Could you please clarify which file you were referring to? Is it the few-shot configuration? I’ve already completed that step, but I’ve noticed that the model still detects base classes when images contain elements representing those classes.

I would like to detect objects in my images that don’t have XML files or annotations available. My goal is to utilize the fine-tuning .pth file for this purpose. Would you be able to assist me with this?