[Bug]: Invalid bounding boxes
dnth opened this issue · comments
What happened?
I ran the Roboflow notebook and found that fastdup flagged some of the bounding boxes invalid.
Here are a few of them
From my understanding, the boxes are flagged as invalid when the bounding boxes go beyond the image. In this dataset, all images are of size 416x416. Clearly none of the bounding boxes above go beyond the image size.
Why are they flagged as invalid boxes in fastdup?
What did you expect to see?
Why the boxes are flagged as invalid.
What version of fastdup were you runnning on?
1.38
What version of Python were you running on?
Python 3.10
Operating System
Ubuntu 22.04
Reproduction steps
Run the Roboflow notebook.
Relevant log output
No response
Attach a screenshot [Optional]
No response
Contact Details [Optional]
No response
@dnth please run with verbose=1 and send me the full trace thanks
Hi @dnth I see the folloing
2023-09-12 13:48:13 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/untitled-2-1-_jpg.rf.2cc0804ad0a8d457291b39d72b32ea1f.jpg 0 265 0 61, bounding box is too small, 9 10 skipping.
2023-09-12 13:48:15 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/IMG_1282_jpg.rf.e5d454a356160bd444b71e19c9350f72.jpg 0 368 0 47, bounding box is too small, 3 10 skipping.
2023-09-12 13:48:15 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/IMG_1282_jpg.rf.f669ecf808bd1721e724ef14c7ede206.jpg 395 406 395 9, bounding box is too small, 21 10 skipping.
2023-09-12 13:48:30 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/Almond_13-Original-Size-_jpg.rf.61e38e0da9a2aa99b20ff801c44a47dd.jpg 409 285 409 40, bounding box is too small, 6 10 skipping.
2023-09-12 13:48:45 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/carrot_93_jpg.rf.a242faf4608f26c230f476b5feb9f27c.jpg 0 324 0 45, bounding box is too small, 2 10 skipping.
2023-09-12 13:49:16 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/pistachios-1570x1047_jpg.rf.9b4218a77ea0530cac6db0abb3491392.jpg 0 295 0 77, bounding box is too small, 8 10 skipping.
2023-09-12 13:49:32 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/cabbage-green-50lb-case_jpg.rf.9b8d2677a0a8c7c0d4ceb961cc91aaab.jpg 333 346 333 1, bounding box is too small, 2 10 skipping.
2023-09-12 13:49:32 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/cabbage-green-50lb-case_jpg.rf.5459cd9f2b964dc5e84fb29905f95353.jpg 361 384 361 1, bounding box is too small, 2 10 skipping.
2023-09-12 13:49:32 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/cabbage-green-50lb-case_jpg.rf.546e265a2b83cd03e7d5d7c98c2d51ef.jpg 28 331 28 2, bounding box is too small, 1 10 skipping.
2023-09-12 13:49:32 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/cabbage-green-50lb-case_jpg.rf.6887156a700058674a259ce271ea74fe.jpg 40 333 40 2, bounding box is too small, 1 10 skipping.
2023-09-12 13:49:33 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/cabbage-green-50lb-case_jpg.rf.f83f582b75b3ba9ed396004997edc7a2.jpg 364 92 364 2, bounding box is too small, 1 10 skipping.
2023-09-12 13:49:45 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/Pomegranates-shutterstock_152360780_jpg.rf.8c5c9586aaa67739a99211a76fc21359.jpg 409 39 409 30, bounding box is too small, 6 10 skipping.
2023-09-12 13:49:47 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/Pomegranates-shutterstock_152360780_jpg.rf.71c53b5feee5f6677bddb176380f2b6c.jpg 44 0 44 9, bounding box is too small, 30 10 skipping.
2023-09-12 13:50:17 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/OIP-24-_jpg.rf.23fc31cbd35c1de823b99968a19e6d2c.jpg 411 290 411 99, bounding box is too small, 4 10 skipping.
2023-09-12 13:50:19 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/90d098b212fc9b6d28ae069e150f8f75_jpg.rf.fadbb69263e413a39ed153caf9d23e0c.jpg 412 259 412 75, bounding box is too small, 4 10 skipping.
2023-09-12 13:50:39 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/Potato_38_jpg.rf.2ba6902291f27de4faa2b53291f7a873.jpg 0 197 0 46, bounding box is too small, 9 10 skipping.
2023-09-12 13:50:39 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/Potato_38_jpg.rf.3315d9c2535e1b2eb1e1b9045e4979f1.jpg 0 196 0 46, bounding box is too small, 7 10 skipping.
2023-09-12 13:50:54 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/Single-Blackberries_28_jpg.rf.465b276617c3780db8a717ce58c49ba9.jpg 122 0 122 9, bounding box is too small, 38 10 skipping.
2023-09-12 13:50:55 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/Egg-images_43_jpg.rf.28d73a7161b833d02ef40017086a9948.jpg 195 27 195 1, bounding box is too small, 1 10 skipping.
2023-09-12 13:50:56 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/Egg-images_43_jpg.rf.0c240a42a1d5a122bf0df56c6737ce1f.jpg 195 394 195 1, bounding box is too small, 1 10 skipping.
2023-09-12 13:50:56 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/Egg-images_43_jpg.rf.0d7f52ff01b12da4a0e1860cef3755fa.jpg 190 403 190 1, bounding box is too small, 1 10 skipping.
2023-09-12 13:50:57 [DEBUG] Found bad bounding box2 for image DASH-DIET-101-4/train/Egg-images_43_jpg.rf.76de2b114b9b114cfd8262ba6a9b6697.jpg 26 220 26 1, bounding box is too small, 1 10 skipping.
~/visual_database/cxx guy_fix *15 ?325 grep "bounding box is too small" ~/Downloads/output.txt | wc ✔ 20:56:02
22 521 4606
There are 22 bounding boxes where the bb is 10 or less. If you like to process all bounding boxes despite they are too small, you can run with augmentation_additive_margin=10.
p.s.
let me know if there is a bug with the invalid instances display.
If anyone encounters this issue in the future, it might be the case that the filename is too long. Try to shorten the filename and see if the issue persists.