search

wangchen1801 / FPD

Official code of the paper "Fine-Grained Prototypes Distillation for Few-Shot Object Detection (AAAI 2024)"
https://arxiv.org/pdf/2401.07629.pdf
27 stars 4 forks source link

transformers.py #9

Closed yefeng23323 closed 4 months ago

yefeng23323 commented 6 months ago

Hello, your code has the file transformers.py, may I ask what it does specifically?I can run the code by deleting this file.

wangchen1801 commented 6 months ago

Hello yefeng, transforms.py contains some image preprocessing methods. See the file below for more functions. 

mmfewshot/mmfewshot/detection/datasets/pipelines/transforms.py

For example, CropResizeInstance defines how to obtain a fixed-size image patch (224, 224) from the original support image. CropResizeInstanceByRatio here does basically the same effect. It is only used in MS COCO dataset and doesn't make a significant difference. So it's ok to delete it and use the CropResizeInstance function as a replacement.

yefeng23323 commented 6 months ago

Hello, it's just that I didn't find a place to use transforms.py in the source code, so you can still train and test normally (using only VOC datasets) by removing it.What do you mean by using CropResizeInstance instead?

wangchen1801 commented 6 months ago

Hello, transforms.py is only used while training on COCO dataset. Your understanding is correct, we don't need to modify the code. Best~