search

ucbdrive / few-shot-object-detection

Implementations of few-shot object detection benchmarks
Apache License 2.0
1.08k stars 225 forks source link

train on a custom dataset #130

Closed ElifErarslan1 closed 2 years ago

ElifErarslan1 commented 2 years ago

Hello, Could you please provide information about how to use FsDet on a custom dataset with VOC format?

AISoltani commented 2 years ago

Hello, Could you please provide information about how to use FsDet on a custom dataset with VOC format?

hello, first you need to edit builtin and meta_builtin and also based on your dataset that like pascal or coco, you need to edit prepare_voc or coco too...run prepare to generate seed and not need to run both first file.

thomasehuang commented 2 years ago

As @AISoltani said, the protocol for creating a custom dataset is to create your dataset file in fsdet/data and update fsdet/data/builtin.py to register dataset paths and fsdet/data/builtin_meta.py to include meta information.

Since you said your dataset is in VOC format, you should just be able to use the VOC dataset file and only need to update fsdet/data/builtin.py and potentially fsdet/data/builtin_meta.py. Refer to the VOC portions of the code as example.

thomasehuang commented 2 years ago

Added more detailed instructions for custom datasets in the README here. Hope that helps!