fiveai / detection_calibration

[ECCV 2024 - Oral Presentation] Python library that provides tools for calibrating object detectors and evaluating them
https://arxiv.org/abs/2405.20459
Other
24 stars 3 forks source link

Query regarding detection(.bbox.json) generation #1

Closed iabh1shekbasu closed 1 month ago

iabh1shekbasu commented 1 month ago

Hi, Thank you for your amazing work. Can you please let me know how can these files be generated?

detections ├── deformable_detr_cityscapes_minival.bbox.json ├── deformable_detr_cityscapes_minitest.bbox.json ├── deformable_detr_coco_minival.bbox.json ├── deformable_detr_coco_minitest.bbox.json ├── ...

selimkuzucu commented 1 month ago

Hi @iabh1shekbasu, thank you for your interest in our work. The files you have mentioned are the final detection files in the COCO format and can be extracted separately for the individual models. The way to generate them would depend on the framework of your choice.

For our experiments, we utilized the mmdetection framework, and specifically its D-DETR implementation for the files you have asked for. To reproduce those files, you need to set the --format_only flag to True and specify a corresponding --outfile_prefix. The way to that would again depend on the version of mmdetection that you have installed, it can either be done through passing the aforementioned flags as command line parameters (for mmdetection 2.X) or through setting the relevant flags within the evaluators (for mmdetection 3.X). For a more detailed explanation, feel free to see "Dataset and evaluator config" Section under the mmdetection user guides

For other established frameworks such as detectron2, please take a look at their respective repositories as I am not fully familiar with them. (Although his may be helpful for detectron2)

Again, thanks for your interest in our work and do let us know if you need any help with anything else.

iabh1shekbasu commented 1 month ago

Thank you very much for your kind response. I was successfully able to generate the detection files.