search

javiribera / locating-objects-without-bboxes

PyTorch code for "Locating objects without bounding boxes" - Loss function and trained models
Other
249 stars 52 forks source link

how to test images without gt #29

Closed JiangAndy closed 4 years ago

JiangAndy commented 4 years ago

After I have trained my model, how can I test my model, I only have images without gt(csv or xml).Thank you!

JiangAndy commented 4 years ago

I have solved it! I make a gt.csv,like following: filename,count,locations img1.png,0,"[()]" img2.png,0,"[()]"

then I add some limits in data.py (class CSVDataset:), add a line if self.there_is_gt in where 'locations' or 'count' are mentioned, because we don't have gts about there images.

1 image

javiribera commented 4 years ago

Thank you! I believe I solved it with commit d8485608c2625d675e61dfd692da675e8dde4225

For the record, this situation was intended to be solved with the --evaluate option (actually NOT using --evaluate). So the idea is if you don't have grounf truth, then you cannot evaluate metrics but you can store the output. But I confirm that not using --evaluate when not having any GT file was crashing. Now that commit fixes it.