Closed Stasolet closed 1 year ago
Hi! thanks for your contribution!, great first issue!
Hi @Stasolet, thanks for reporting this issue.
Regarding the first issue of compability with numpy >1.20.0: There is really not mush we can do here as long as it does not get fixed in the official repository. I know there are open PRs to fix it, but since the repository is not really maintained anymore I do really not have high hopes. I recommend install the fork on pypi as that is maintained and supports numpy > 1.20.0.
Regarding the confusion about the documentation:
You are completely right that the language in the note is not explicit enough. Essentially, we only detect if a package named pycocotools
is installed, not where it is installed from. Therefore, we support both the official implementation and fork found on pypi because both installs as pycocotools
. But this is not clear.
I fixed the note in PR https://github.com/Lightning-AI/torchmetrics/pull/2034 to make it more clear what we actually mean where we also add support for another backend: https://github.com/MiXaiLL76/faster_coco_eval
🐛 Bug
To Reproduce
Code sample used from documentation:
```python from torch import tensor from torchmetrics.detection import MeanAveragePrecision preds = [ dict( boxes=tensor([[258.0, 41.0, 606.0, 285.0]]), scores=tensor([0.536]), labels=tensor([0]), ) ] target = [ dict( boxes=tensor([[214.0, 41.0, 562.0, 285.0]]), labels=tensor([0]), ) ] metric = MeanAveragePrecision(iou_type="bbox") metric.update(preds, target) from pprint import pprint pprint(metric.compute()) ```Environment
Additional context
In documentation for map I found "This metric utilizes the official pycocotools implementation as its backend". But there is a small confusing point: the official repository "pycocotools" and "pycocotools" package in pypi are different things. Which official pycocotools package was meant in the documentation?