Detectron: Non-pdf output

[sh0gg0th]

Expected results

Is there a way to output raw detection results? Ideally, I would like to be able to produce a JSON that would include for each object class the count of detected objects and their share of the total image area - where should I start looking? Thanks.

Actual results

What did you observe instead?

Detailed steps to reproduce

E.g.:

The command that you ran

System information

• Operating system: ?
• Compiler version: ?
• CUDA version: ?
• cuDNN version: ?
• NVIDIA driver version: ?
• GPU models (for all devices if they are not all the same): ?
• PYTHONPATH environment variable: ?
• python --version output: ?
• Anything else that seems relevant: ?

[fabito]

You can create a new script based on infer_simple.py but instead of generating a PDF ( vis_utils.vis_one_image() ) you could just print the data you need to a file.

[kampelmuehler]

@sh0gg0th after an inference run a file detections.pkl is written to the output folder, which contains all the raw detections.

[luisriera]

Hello there,

@kampelmuehler, is there a way to activate detections.pkl from infer_simple.py or I have to make my own script? By default, the infer_simple.py I have does not write a detections.pkl file.

[kampelmuehler]

detections.pkl is created automatically whenever you run inference. At least as of February 18.

[JavierClearImageAI]

detections.pkl is created automatically whenever you run inference. At least as of February 18.

do you know where exactly? not found in tmp