PQ value for each validation image

[sumanth2002629]

Is there any way in which I can get the PQ value for each validation image of the cityscapes dataset while running the evaluation??

When I run the evaluation, I get the combined metrics. Where do I see the metrics of each validation image?

[praeclarumjj3]

Hi @sumanth2002629, there should be a way to obtain that by using the following: panoptic_evaluation.py and pq_compute function.

(Note that the COCOPanopticEvaluator is used with cityscapes).

You can obtain the per-image PQ scores inside the pq_compute function by storing the PQ score for each image in the loop here. You can then return the per-image-PQ scores from pq_compute to the COCOPanopticEvaluator.

Of course, you might have to make a few other changes to make it work.

[praeclarumjj3]

Closing, feel free to reopen.

[HaonanZhao1]

Hi. Is there any way to find the code that transfers the prediction results to the PQ evaluator on Cityscapes dataset?