Could we use custom dataset in voc format for object detection?

[ctr-226]

Could we use custom dataset in voc format for object detection?
I noticed that you use voc2012 for segmentation and coco for detection.
I am wondering if I can do target detection with a voc format dataset or do I have to use coco format?
Thanks!

[ctr-226]

By the way, I am trying to use SSD detect model. What should I do to run L-CRP with my own model on a custom dataset?
I am new to object detection or segmentation. I'm sorry if my question was weird.

[maxdreyer]

Hi!
The code base allows to add custom datasets and models. However, the dataset and model class should follow certain rules in order to work (e.g., correct output format).
Make sure to add the dataset and model in the datasets/__init__.py and models/__init__.py files.
Further, also add the CRP/LRP config in utils/crp_configs.py. Note, that LRP (canonizer/composite) must be compatible with your model.

If you want to use VOC and SSD, I would suggest using code from https://github.com/lufficc/SSD, where I took the COCO dataset and SSD model class. I would assume this is the easiest to implement.
If you have further questions/problems, please reach out to maximilian.dreyer@hhi.fraunhofer.de.

[ctr-226]

Hi! The code base allows to add custom datasets and models. However, the dataset and model class should follow certain rules in order to work (e.g., correct output format). Make sure to add the dataset and model in the datasets/__init__.py and models/__init__.py files. Further, also add the CRP/LRP config in utils/crp_configs.py. Note, that LRP (canonizer/composite) must be compatible with your model.

If you want to use VOC and SSD, I would suggest using code from https://github.com/lufficc/SSD, where I took the COCO dataset and SSD model class. I would assume this is the easiest to implement. If you have further questions/problems, please reach out to maximilian.dreyer@hhi.fraunhofer.de.

Thank you for answering. I am working on that.