Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.
And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.
Can't reproduce the culane results
See original GitHub issueI used the pre-trained weights and python main.py test --exp_name laneatt_r18/34/122_culane.
The results do not match with the paper.
laneatt_r18_culane
{'TP': 61583, 'FP': 25973, 'FN': 31363, 'Precision': 0.7033555667230116, 'Recall': 0.6625675123189809, 'F1': 0.68223525501102481}
laneatt_r34_culane
{'TP': 61954, 'FP': 28336, 'FN': 30992, 'Precision': 0.6861667958799424, 'Recall': 0.666559077313709, 'F1': 0.6762208299679102}
laneatt_r122_culane
{'TP': 60938, 'FP': 25244, 'FN': 32008, 'Precision': 0.7070850061497761, 'Recall': 0.6556279990532137, 'F1': 0.6803849761064713}
BTW tusimple results perfectly match.
Issue Analytics
- State:
- Created 2 years ago
- Comments:6 (3 by maintainers)
Top Results From Across the Web
Top Related Medium Post
Top Related StackOverflow Question
Troubleshoot Live Code
Top Related Reddit Thread
Top Related Hackernoon Post
Top Related Tweet
Top Related Dev.to Post
Top Related Hashnode Post
Thx ! I redownloaded the CULane testset and re-evaluate the pretrained model. r18:
TP: 72257, FP: 15299, FN: 32629, Precision: 0.8253, Recall: 0.6889, F1: 0.7509; r34:TP: 74676, FP: 15281, FN: 30210, Precision: 0.8301, Recall: 0.7120, F1: 0.7665; r122:TP: 73579, FP: 12603, FN: 31307, Precision: 0.8538, Recall: 0.7015, F1: 0.7702;The correct number is 104886. The number of tested images should be 34680.