Add lane classification feature
zillur-av opened this issue · comments
I have a dataset in tusimple format but with lane class labels like white-dashed, white-solid, yellow-solid. Do you have any suggestions on how to modify the network heads and loss function?
@zillur-av
simpley add a 2-way classification head for every lane should be ok. Or you can directly flatten the CNN feature, and feed it to an MLP to achieve this. For loss function, cross-entropy should be ok.
@zillur-av simpley add a 2-way classification head for every lane should be ok. Or you can directly flatten the CNN feature, and feed it to an MLP to achieve this. For loss function, cross-entropy should be ok.
Appreciate your reply. I went through the entire paper which you made easy to follow. I am interested in that 2-way classification head, which means I want to add a new head along with your detection head. Your current network is backbone - conv2D(8 channel) - FC1+relu (2048) - FC2(4*101*56) where the last FC is transformed into (4*101*56) tensor for lane detection.
If I add 4 heads for a maximum of 4 lanes, the total would be 5 heads. Each classification head will give lane classes or the background. Is that the idea? You converted each image label to a 4*56 matrix where matrix element=100 means it is the background. If I understand correctly, it means it is dealing with just lane or background, it does not care for lane category. I guess I can not use these kinds of labels for classification.
@zillur-av
Sorry, the 2-way classification is a mistake. If you have white-dashed, white-solid, and yellow-solid, it should be a 3-way classification.
For network structure:
If you have 4 lanes, you can use 4*3 (suppose there are 3 lane categories) digits to represent the lane classficition results. In this way, you can have two ways:
- add the lane classification head to the conv features, it would be like:
backbone - conv2D(8 channel) |- FC1+relu (2048) - FC2(4*101*56) |- FC1+relu (2048) - FC2(4*3) - or share the first FC layer, it would be like:
backbone - conv2D(8 channel) - FC1+relu (2048) |- FC2(4*101*56) |- FC2(4*3)
I don't know which one is better, but I think both should work.
For the lane category label:
Yes, the 4*56 matrix where matrix element=100 is only used for localization and background. You need extra label to finish the lane category prediction.
Thanks. I followed your first method. The classification head output is (batch size, num_lanes, num_classes) which is (4, 4, 3). The target shape is (batch size, num_lanes) where each lane index could be 0 to 2 like
tensor([[0, 0, 2, 1],
[0, 1, 1, 1],
[0, 2, 2, 1],
[1, 1, 2, 1]])
It throws an error for torch.nn.CrossEntropyLoss() for shape mismatch. So, I changed classification head output to (4,3,4) and it is working. But after training for around 40 epochs, I see loss is not decreasing. I added the new classification loss with the existing detection loss and send the total loss for optimization.
@zillur-av
First, (4,3,4) is the correct format for CEloss.
What's the loss coefficient? Could you please post the value of lane category loss and other losses?
Here is a log file after deleting a few intemediate rows: det_loss is your model loss which is decreasing just fine and cat_loss is the new head loss. I added both of them and then send them to loss.backward(). I am suspecting underfitting problem.
2023-04-18 11:17:20,042 - lanedet.utils.recorder - INFO - Build train loader...
2023-04-18 11:17:20,043 - lanedet.datasets.base_dataset - INFO - Loading TuSimple annotations...
2023-04-18 11:17:20,473 - lanedet.utils.recorder - INFO - Start training...
2023-04-18 11:17:22,177 - lanedet.utils.recorder - INFO - epoch: 0 step: 1 lr: 0.0000 det_loss: 4.5392 cat_loss: 2.0801 data: 0.9427 batch: 1.6959 eta: 11:22:34
2023-04-18 11:17:29,473 - lanedet.utils.recorder - INFO - epoch: 0 step: 101 lr: 0.0005 det_loss: 4.4104 cat_loss: 2.0304 data: 0.0291 batch: 0.0722 eta: 0:35:37
2023-04-18 11:17:37,006 - lanedet.utils.recorder - INFO - epoch: 0 step: 201 lr: 0.0010 det_loss: 1.8694 cat_loss: 1.6869 data: 0.0295 batch: 0.0770 eta: 0:32:47
2023-04-18 11:17:44,332 - lanedet.utils.recorder - INFO - epoch: 0 step: 301 lr: 0.0015 det_loss: 1.7411 cat_loss: 1.6847 data: 0.0271 batch: 0.0684 eta: 0:31:29
2023-04-18 11:17:52,153 - lanedet.utils.recorder - INFO - epoch: 0 step: 401 lr: 0.0020 det_loss: 1.6074 cat_loss: 1.6819 data: 0.0359 batch: 0.0814 eta: 0:31:14
2023-04-18 11:17:59,485 - lanedet.utils.recorder - INFO - epoch: 0 step: 501 lr: 0.0025 det_loss: 1.5437 cat_loss: 1.6841 data: 0.0304 batch: 0.0716 eta: 0:30:40
2023-04-18 11:18:06,697 - lanedet.utils.recorder - INFO - epoch: 0 step: 601 lr: 0.0029 det_loss: 1.4303 cat_loss: 1.7120 data: 0.0306 batch: 0.0721 eta: 0:30:10
2023-04-18 11:18:13,765 - lanedet.utils.recorder - INFO - epoch: 0 step: 701 lr: 0.0034 det_loss: 1.4222 cat_loss: 1.6864 data: 0.0312 batch: 0.0791 eta: 0:29:41
2023-04-18 11:18:20,882 - lanedet.utils.recorder - INFO - epoch: 0 step: 801 lr: 0.0039 det_loss: 1.3905 cat_loss: 1.6656 data: 0.0272 batch: 0.0667 eta: 0:29:20
2023-04-18 11:18:21,075 - lanedet.utils.recorder - INFO - epoch: 0 step: 804 lr: 0.0039 det_loss: 1.4081 cat_loss: 1.6694 data: 0.0269 batch: 0.0650 eta: 0:29:19
2023-04-18 11:18:21,229 - lanedet.datasets.base_dataset - INFO - Loading TuSimple annotations...
2023-04-18 11:18:31,161 - lanedet.utils.recorder - INFO - Detection: 0.4071101916376303 classification accuracy: 0.3313106796116506
2023-04-18 11:18:33,548 - lanedet.utils.recorder - INFO - Best detection metric: 0.4071101916376303 Best classification metric: 0.3313106796116506
2023-04-18 11:18:34,530 - lanedet.utils.recorder - INFO - epoch: 1 step: 805 lr: 0.0039 det_loss: 1.4011 cat_loss: 1.6820 data: 0.0698 batch: 0.1097 eta: 0:29:44
2023-04-18 11:18:41,829 - lanedet.utils.recorder - INFO - epoch: 1 step: 905 lr: 0.0044 det_loss: 1.3765 cat_loss: 1.6756 data: 0.0293 batch: 0.0699 eta: 0:29:27
2023-04-18 11:18:48,979 - lanedet.utils.recorder - INFO - epoch: 1 step: 1005 lr: 0.0048 det_loss: 1.3803 cat_loss: 1.6696 data: 0.0280 batch: 0.0699 eta: 0:29:09
2023-04-18 11:18:56,069 - lanedet.utils.recorder - INFO - epoch: 1 step: 1105 lr: 0.0053 det_loss: 1.3570 cat_loss: 1.6635 data: 0.0284 batch: 0.0712 eta: 0:28:52
2023-04-18 11:19:03,257 - lanedet.utils.recorder - INFO - epoch: 1 step: 1205 lr: 0.0058 det_loss: 1.2873 cat_loss: 1.6567 data: 0.0290 batch: 0.0719 eta: 0:28:38
2023-04-18 11:19:10,378 - lanedet.utils.recorder - INFO - epoch: 1 step: 1305 lr: 0.0062 det_loss: 1.3041 cat_loss: 1.5790 data: 0.0282 batch: 0.0714 eta: 0:28:24
2023-04-18 11:19:17,622 - lanedet.utils.recorder - INFO - epoch: 1 step: 1405 lr: 0.0067 det_loss: 1.2667 cat_loss: 1.5938 data: 0.0328 batch: 0.0773 eta: 0:28:13
2023-04-18 11:19:24,978 - lanedet.utils.recorder - INFO - epoch: 1 step: 1505 lr: 0.0071 det_loss: 1.2208 cat_loss: 1.5343 data: 0.0266 batch: 0.0706 eta: 0:28:04
2023-04-18 11:19:31,933 - lanedet.utils.recorder - INFO - epoch: 1 step: 1605 lr: 0.0075 det_loss: 1.1481 cat_loss: 1.5654 data: 0.0274 batch: 0.0666 eta: 0:27:50
2023-04-18 11:19:32,121 - lanedet.utils.recorder - INFO - epoch: 1 step: 1608 lr: 0.0076 det_loss: 1.1637 cat_loss: 1.5714 data: 0.0268 batch: 0.0649 eta: 0:27:49
2023-04-18 11:19:41,694 - lanedet.utils.recorder - INFO - Detection: 0.674608013937282 classification accuracy: 0.5841423948220068
2023-04-18 11:19:45,870 - lanedet.utils.recorder - INFO - Best detection metric: 0.674608013937282 Best classification metric: 0.5841423948220068
2023-04-18 11:19:46,857 - lanedet.utils.recorder - INFO - epoch: 2 step: 1609 lr: 0.0076 det_loss: 1.1644 cat_loss: 1.5613 data: 0.0708 batch: 0.1097 eta: 0:28:02
2023-04-18 11:19:54,097 - lanedet.utils.recorder - INFO - epoch: 2 step: 1709 lr: 0.0080 det_loss: 1.1395 cat_loss: 1.5304 data: 0.0316 batch: 0.0691 eta: 0:27:51
2023-04-18 11:20:01,111 - lanedet.utils.recorder - INFO - epoch: 2 step: 1809 lr: 0.0084 det_loss: 1.1515 cat_loss: 1.5235 data: 0.0312 batch: 0.0698 eta: 0:27:38
2023-04-18 11:20:08,107 - lanedet.utils.recorder - INFO - epoch: 2 step: 1909 lr: 0.0089 det_loss: 1.1339 cat_loss: 1.5149 data: 0.0304 batch: 0.0698 eta: 0:27:26
2023-04-18 11:20:15,090 - lanedet.utils.recorder - INFO - epoch: 2 step: 2009 lr: 0.0093 det_loss: 1.1102 cat_loss: 1.5424 data: 0.0278 batch: 0.0701 eta: 0:27:14
2023-04-18 11:20:22,131 - lanedet.utils.recorder - INFO - epoch: 2 step: 2109 lr: 0.0097 det_loss: 1.0651 cat_loss: 1.5117 data: 0.0296 batch: 0.0724 eta: 0:27:03
2023-04-18 11:20:29,156 - lanedet.utils.recorder - INFO - epoch: 2 step: 2209 lr: 0.0101 det_loss: 1.0416 cat_loss: 1.4851 data: 0.0284 batch: 0.0702 eta: 0:26:52
2023-04-18 11:20:36,150 - lanedet.utils.recorder - INFO - epoch: 2 step: 2309 lr: 0.0106 det_loss: 1.0762 cat_loss: 1.4834 data: 0.0297 batch: 0.0709 eta: 0:26:42
2023-04-18 11:20:42,997 - lanedet.utils.recorder - INFO - epoch: 2 step: 2409 lr: 0.0110 det_loss: 1.0327 cat_loss: 1.4152 data: 0.0266 batch: 0.0657 eta: 0:26:30
2023-04-18 11:20:43,189 - lanedet.utils.recorder - INFO - epoch: 2 step: 2412 lr: 0.0110 det_loss: 1.0460 cat_loss: 1.4174 data: 0.0273 batch: 0.0660 eta: 0:26:29
2023-04-18 11:20:53,235 - lanedet.utils.recorder - INFO - Detection: 0.812594367015098 classification accuracy: 0.7176375404530744
2023-04-18 11:20:57,354 - lanedet.utils.recorder - INFO - Best detection metric: 0.812594367015098 Best classification metric: 0.7176375404530744
2023-04-18 11:20:58,405 - lanedet.utils.recorder - INFO - epoch: 3 step: 2413 lr: 0.0110 det_loss: 1.0359 cat_loss: 1.4163 data: 0.0728 batch: 0.1143 eta: 0:26:38
2023-04-18 11:21:05,628 - lanedet.utils.recorder - INFO - epoch: 3 step: 2513 lr: 0.0114 det_loss: 1.0276 cat_loss: 1.4162 data: 0.0296 batch: 0.0685 eta: 0:26:29
2023-04-18 11:21:12,808 - lanedet.utils.recorder - INFO - epoch: 3 step: 2613 lr: 0.0118 det_loss: 1.0263 cat_loss: 1.5376 data: 0.0270 batch: 0.0712 eta: 0:26:21
2023-04-18 11:21:19,903 - lanedet.utils.recorder - INFO - epoch: 3 step: 2713 lr: 0.0122 det_loss: 0.9982 cat_loss: 1.4510 data: 0.0312 batch: 0.0699 eta: 0:26:11
2023-04-18 11:21:27,045 - lanedet.utils.recorder - INFO - epoch: 3 step: 2813 lr: 0.0126 det_loss: 0.9795 cat_loss: 1.4290 data: 0.0291 batch: 0.0727 eta: 0:26:03
2023-04-18 11:21:34,234 - lanedet.utils.recorder - INFO - epoch: 3 step: 2913 lr: 0.0130 det_loss: 0.9795 cat_loss: 1.4036 data: 0.0313 batch: 0.0737 eta: 0:25:54
2023-04-18 11:21:41,463 - lanedet.utils.recorder - INFO - epoch: 3 step: 3013 lr: 0.0134 det_loss: 0.9330 cat_loss: 1.3765 data: 0.0304 batch: 0.0729 eta: 0:25:46
2023-04-18 11:22:06,095 - lanedet.utils.recorder - INFO - Detection: 0.8233231707317069 classification accuracy: 0.7419093851132684
2023-04-18 11:22:10,276 - lanedet.utils.recorder - INFO - Best detection metric: 0.8233231707317069 Best classification metric: 0.7419093851132684
2023-04-18 11:22:11,294 - lanedet.utils.recorder - INFO - epoch: 4 step: 3217 lr: 0.0141 det_loss: 0.9166 cat_loss: 1.4365 data: 0.0718 batch: 0.1105 eta: 0:25:34
2023-04-18 11:22:19,077 - lanedet.utils.recorder - INFO - epoch: 4 step: 3317 lr: 0.0145 det_loss: 0.9526 cat_loss: 1.4885 data: 0.0288 batch: 0.0745 eta: 0:25:29
2023-04-18 11:22:26,290 - lanedet.utils.recorder - INFO - epoch: 4 step: 3417 lr: 0.0149 det_loss: 0.9538 cat_loss: 1.4566 data: 0.0280 batch: 0.0712 eta: 0:25:21
2023-04-18 11:22:33,857 - lanedet.utils.recorder - INFO - epoch: 4 step: 3517 lr: 0.0153 det_loss: 0.9202 cat_loss: 1.4404 data: 0.0327 batch: 0.0850 eta: 0:25:15
2023-04-18 11:22:41,187 - lanedet.utils.recorder - INFO - epoch: 4 step: 3617 lr: 0.0156 det_loss: 0.9457 cat_loss: 1.4739 data: 0.0285 batch: 0.0747 eta: 0:25:08
2023-04-18 11:22:48,578 - lanedet.utils.recorder - INFO - epoch: 4 step: 3717 lr: 0.0160 det_loss: 0.8939 cat_loss: 1.3964 data: 0.0297 batch: 0.0780 eta: 0:25:01
2023-04-18 11:23:21,911 - lanedet.utils.recorder - INFO - Detection: 0.8709240708478517 classification accuracy: 0.7463592233009712
2023-04-18 11:23:26,146 - lanedet.utils.recorder - INFO - Best detection metric: 0.8709240708478517 Best classification metric: 0.7463592233009712
2023-04-18 11:23:27,317 - lanedet.utils.recorder - INFO - epoch: 5 step: 4021 lr: 0.0171 det_loss: 0.8586 cat_loss: 1.4136 data: 0.0798 batch: 0.1207 eta: 0:24:44
2023-04-18 11:23:35,092 - lanedet.utils.recorder - INFO - epoch: 5 step: 4121 lr: 0.0174 det_loss: 0.8674 cat_loss: 1.3714 data: 0.0287 batch: 0.0737 eta: 0:24:38
2023-04-18 11:23:42,829 - lanedet.utils.recorder - INFO - epoch: 5 step: 4221 lr: 0.0178 det_loss: 0.8170 cat_loss: 1.4132 data: 0.0294 batch: 0.0744 eta: 0:24:33
2023-04-18 11:23:50,305 - lanedet.utils.recorder - INFO - epoch: 5 step: 4321 lr: 0.0181 det_loss: 0.8060 cat_loss: 1.3951 data: 0.0278 batch: 0.0703 eta: 0:24:26
2023-04-18 11:23:57,530 - lanedet.utils.recorder - INFO - epoch: 5 step: 4421 lr: 0.0184 det_loss: 0.8374 cat_loss: 1.5034 data: 0.0288 batch: 0.0730 eta: 0:24:18
2023-04-18 11:24:04,944 - lanedet.utils.recorder - INFO - epoch: 5 step: 4521 lr: 0.0188 det_loss: 0.8961 cat_loss: 1.5188 data: 0.0299 batch: 0.0760 eta: 0:24:10
2023-04-18 11:24:37,735 - lanedet.utils.recorder - INFO - Detection: 0.8464975319396056 classification accuracy: 0.6427993527508089
2023-04-18 11:24:37,735 - lanedet.utils.recorder - INFO - Best detection metric: 0.8709240708478517 Best classification metric: 0.7463592233009712
2023-04-18 11:24:38,943 - lanedet.utils.recorder - INFO - epoch: 6 step: 4825 lr: 0.0197 det_loss: 0.8527 cat_loss: 1.4535 data: 0.0824 batch: 0.1246 eta: 0:23:52
2023-04-18 11:24:46,625 - lanedet.utils.recorder - INFO - epoch: 6 step: 4925 lr: 0.0201 det_loss: 0.8356 cat_loss: 1.4509 data: 0.0276 batch: 0.0727 eta: 0:23:46
2023-04-18 11:24:54,401 - lanedet.utils.recorder - INFO - epoch: 6 step: 5025 lr: 0.0203 det_loss: 0.8907 cat_loss: 1.5206 data: 0.0278 batch: 0.0735 eta: 0:23:40
2023-04-18 11:25:01,444 - lanedet.utils.recorder - INFO - epoch: 6 step: 5125 lr: 0.0202 det_loss: 0.8552 cat_loss: 1.4906 data: 0.0267 batch: 0.0698 eta: 0:23:31
2023-04-18 11:25:08,701 - lanedet.utils.recorder - INFO - epoch: 6 step: 5225 lr: 0.0201 det_loss: 0.8129 cat_loss: 1.4638 data: 0.0299 batch: 0.0727 eta: 0:23:23
2023-04-18 11:25:15,864 - lanedet.utils.recorder - INFO - epoch: 6 step: 5325 lr: 0.0200 det_loss: 0.7955 cat_loss: 1.4348 data: 0.0254 batch: 0.0729 eta: 0:23:15
2023-04-18 11:25:47,598 - lanedet.utils.recorder - INFO - Detection: 0.8619446864111502 classification accuracy: 0.6917475728155341
2023-04-18 11:25:47,598 - lanedet.utils.recorder - INFO - Best detection metric: 0.8709240708478517 Best classification metric: 0.7463592233009712
2023-04-18 11:25:48,793 - lanedet.utils.recorder - INFO - epoch: 7 step: 5629 lr: 0.0197 det_loss: 0.8069 cat_loss: 1.5202 data: 0.0797 batch: 0.1232 eta: 0:22:53
2023-04-18 11:25:56,512 - lanedet.utils.recorder - INFO - epoch: 7 step: 5729 lr: 0.0196 det_loss: 0.7902 cat_loss: 1.4698 data: 0.0301 batch: 0.0749 eta: 0:22:46
2023-04-18 11:26:03,668 - lanedet.utils.recorder - INFO - epoch: 7 step: 5829 lr: 0.0195 det_loss: 0.8076 cat_loss: 1.4892 data: 0.0270 batch: 0.0697 eta: 0:22:38
2023-04-18 11:26:10,839 - lanedet.utils.recorder - INFO - epoch: 7 step: 5929 lr: 0.0194 det_loss: 0.7473 cat_loss: 1.4581 data: 0.0312 batch: 0.0736 eta: 0:22:30
2023-04-18 11:26:18,142 - lanedet.utils.recorder - INFO - epoch: 7 step: 6029 lr: 0.0193 det_loss: 0.7266 cat_loss: 1.4094 data: 0.0283 batch: 0.0700 eta: 0:22:22
2023-04-18 11:26:25,168 - lanedet.utils.recorder - INFO - epoch: 7 step: 6129 lr: 0.0192 det_loss: 0.7156 cat_loss: 1.3987 data: 0.0296 batch: 0.0715 eta: 0:22:13
2023-04-18 11:26:56,646 - lanedet.utils.recorder - INFO - Detection: 0.8931257259001162 classification accuracy: 0.669498381877023
2023-04-18 11:27:00,747 - lanedet.utils.recorder - INFO - Best detection metric: 0.8931257259001162 Best classification metric: 0.7463592233009712
2023-04-18 11:27:01,875 - lanedet.utils.recorder - INFO - epoch: 8 step: 6433 lr: 0.0189 det_loss: 0.6961 cat_loss: 1.4728 data: 0.0743 batch: 0.1163 eta: 0:21:50
2023-04-18 11:27:09,257 - lanedet.utils.recorder - INFO - epoch: 8 step: 6533 lr: 0.0188 det_loss: 0.7284 cat_loss: 1.4751 data: 0.0300 batch: 0.0714 eta: 0:21:43
2023-04-18 11:27:16,406 - lanedet.utils.recorder - INFO - epoch: 8 step: 6633 lr: 0.0187 det_loss: 0.8478 cat_loss: 1.5592 data: 0.0290 batch: 0.0708 eta: 0:21:35
2023-04-18 11:27:23,712 - lanedet.utils.recorder - INFO - epoch: 8 step: 6733 lr: 0.0186 det_loss: 0.7283 cat_loss: 1.5008 data: 0.0278 batch: 0.0708 eta: 0:21:27
2023-04-18 11:27:30,915 - lanedet.utils.recorder - INFO - epoch: 8 step: 6833 lr: 0.0185 det_loss: 0.7426 cat_loss: 1.4399 data: 0.0275 batch: 0.0717 eta: 0:21:19
2023-04-18 11:27:38,205 - lanedet.utils.recorder - INFO - epoch: 8 step: 6933 lr: 0.0184 det_loss: 0.7632 cat_loss: 1.4697 data: 0.0279 batch: 0.0792 eta: 0:21:12
2023-04-18 11:28:10,450 - lanedet.utils.recorder - INFO - Detection: 0.8673562717770037 classification accuracy: 0.6605987055016183
2023-04-18 11:28:10,450 - lanedet.utils.recorder - INFO - Best detection metric: 0.8931257259001162 Best classification metric: 0.7463592233009712
2023-04-18 11:28:11,731 - lanedet.utils.recorder - INFO - epoch: 9 step: 7237 lr: 0.0181 det_loss: 0.6763 cat_loss: 1.4755 data: 0.0844 batch: 0.1280 eta: 0:20:50
2023-04-18 11:28:19,209 - lanedet.utils.recorder - INFO - epoch: 9 step: 7337 lr: 0.0180 det_loss: 0.6609 cat_loss: 1.4502 data: 0.0287 batch: 0.0744 eta: 0:20:43
2023-04-18 11:28:26,748 - lanedet.utils.recorder - INFO - epoch: 9 step: 7437 lr: 0.0179 det_loss: 0.7027 cat_loss: 1.5152 data: 0.0268 batch: 0.0707 eta: 0:20:36
2023-04-18 11:28:34,057 - lanedet.utils.recorder - INFO - epoch: 9 step: 7537 lr: 0.0179 det_loss: 0.7075 cat_loss: 1.4915 data: 0.0281 batch: 0.0700 eta: 0:20:28
2023-04-18 11:28:41,490 - lanedet.utils.recorder - INFO - epoch: 9 step: 7637 lr: 0.0178 det_loss: 0.6720 cat_loss: 1.4494 data: 0.0336 batch: 0.0800 eta: 0:20:21
2023-04-18 11:29:22,235 - lanedet.utils.recorder - INFO - Detection: 0.907324332171893 classification accuracy: 0.6808252427184467
2023-04-18 11:29:26,414 - lanedet.utils.recorder - INFO - Best detection metric: 0.907324332171893 Best classification metric: 0.7463592233009712
2023-04-18 11:29:27,399 - lanedet.utils.recorder - INFO - epoch: 10 step: 8041 lr: 0.0174 det_loss: 0.6273 cat_loss: 1.4672 data: 0.0688 batch: 0.1121 eta: 0:19:53
2023-04-18 11:29:35,385 - lanedet.utils.recorder - INFO - epoch: 10 step: 8141 lr: 0.0173 det_loss: 0.6380 cat_loss: 1.4524 data: 0.0305 batch: 0.0764 eta: 0:19:47
2023-04-18 11:29:42,864 - lanedet.utils.recorder - INFO - epoch: 10 step: 8241 lr: 0.0172 det_loss: 0.6332 cat_loss: 1.4301 data: 0.0270 batch: 0.0741 eta: 0:19:40
2023-04-18 11:29:50,275 - lanedet.utils.recorder - INFO - epoch: 10 step: 8341 lr: 0.0171 det_loss: 0.6841 cat_loss: 1.4801 data: 0.0301 batch: 0.0763 eta: 0:19:32
2023-04-18 11:29:57,926 - lanedet.utils.recorder - INFO - epoch: 10 step: 8441 lr: 0.0170 det_loss: 0.6377 cat_loss: 1.4413 data: 0.0314 batch: 0.0787 eta: 0:19:25
2023-04-18 11:30:05,304 - lanedet.utils.recorder - INFO - epoch: 10 step: 8541 lr: 0.0169 det_loss: 0.6567 cat_loss: 1.4444 data: 0.0285 batch: 0.0737 eta: 0:19:18
2023-04-18 11:30:38,394 - lanedet.utils.recorder - INFO - Detection: 0.9092806329849008 classification accuracy: 0.7148058252427182
2023-04-18 11:30:42,609 - lanedet.utils.recorder - INFO - Best detection metric: 0.9092806329849008 Best classification metric: 0.7463592233009712
2023-04-18 11:30:43,678 - lanedet.utils.recorder - INFO - epoch: 11 step: 8845 lr: 0.0166 det_loss: 0.6206 cat_loss: 1.4078 data: 0.0699 batch: 0.1138 eta: 0:18:57
2023-04-18 11:30:51,407 - lanedet.utils.recorder - INFO - epoch: 11 step: 8945 lr: 0.0165 det_loss: 0.6316 cat_loss: 1.4667 data: 0.0268 batch: 0.0745 eta: 0:18:50
2023-04-18 11:30:59,643 - lanedet.utils.recorder - INFO - epoch: 11 step: 9045 lr: 0.0164 det_loss: 0.6456 cat_loss: 1.4425 data: 0.0303 batch: 0.0761 eta: 0:18:44
2023-04-18 11:31:07,454 - lanedet.utils.recorder - INFO - epoch: 11 step: 9145 lr: 0.0163 det_loss: 0.6501 cat_loss: 1.4470 data: 0.0317 batch: 0.0797 eta: 0:18:37
2023-04-18 11:31:16,234 - lanedet.utils.recorder - INFO - epoch: 11 step: 9245 lr: 0.0162 det_loss: 0.5828 cat_loss: 1.4223 data: 0.0332 batch: 0.0783 eta: 0:18:32
2023-04-18 11:31:57,556 - lanedet.utils.recorder - INFO - Detection: 0.9217116724738679 classification accuracy: 0.682847896440129
2023-04-18 11:32:01,637 - lanedet.utils.recorder - INFO - Best detection metric: 0.9217116724738679 Best classification metric: 0.7463592233009712
2023-04-18 11:32:02,553 - lanedet.utils.recorder - INFO - epoch: 12 step: 9649 lr: 0.0158 det_loss: 0.6029 cat_loss: 1.4384 data: 0.0718 batch: 0.1109 eta: 0:18:04
2023-04-18 11:32:10,298 - lanedet.utils.recorder - INFO - epoch: 12 step: 9749 lr: 0.0157 det_loss: 0.6138 cat_loss: 1.4492 data: 0.0288 batch: 0.0727 eta: 0:17:57
2023-04-18 11:32:17,791 - lanedet.utils.recorder - INFO - epoch: 12 step: 9849 lr: 0.0156 det_loss: 0.6052 cat_loss: 1.4156 data: 0.0327 batch: 0.0751 eta: 0:17:50
2023-04-18 11:32:25,362 - lanedet.utils.recorder - INFO - epoch: 12 step: 9949 lr: 0.0155 det_loss: 0.6100 cat_loss: 1.4157 data: 0.0304 batch: 0.0759 eta: 0:17:42
2023-04-18 11:32:32,969 - lanedet.utils.recorder - INFO - epoch: 12 step: 10049 lr: 0.0154 det_loss: 0.6324 cat_loss: 1.4528 data: 0.0301 batch: 0.0741 eta: 0:17:35
2023-04-18 11:33:13,235 - lanedet.utils.recorder - INFO - Detection: 0.8888138792102206 classification accuracy: 0.738673139158576
2023-04-18 11:33:13,235 - lanedet.utils.recorder - INFO - Best detection metric: 0.9217116724738679 Best classification metric: 0.7463592233009712
2023-04-18 11:33:14,228 - lanedet.utils.recorder - INFO - epoch: 13 step: 10453 lr: 0.0150 det_loss: 0.6152 cat_loss: 1.4634 data: 0.0706 batch: 0.1094 eta: 0:17:05
2023-04-18 11:33:21,736 - lanedet.utils.recorder - INFO - epoch: 13 step: 10553 lr: 0.0149 det_loss: 0.5752 cat_loss: 1.4417 data: 0.0285 batch: 0.0708 eta: 0:16:58
2023-04-18 11:33:29,319 - lanedet.utils.recorder - INFO - epoch: 13 step: 10653 lr: 0.0148 det_loss: 0.6782 cat_loss: 1.4597 data: 0.0305 batch: 0.0769 eta: 0:16:50
2023-04-18 11:33:36,505 - lanedet.utils.recorder - INFO - epoch: 13 step: 10753 lr: 0.0147 det_loss: 0.6082 cat_loss: 1.4522 data: 0.0318 batch: 0.0746 eta: 0:16:43
2023-04-18 11:33:43,896 - lanedet.utils.recorder - INFO - epoch: 13 step: 10853 lr: 0.0146 det_loss: 0.6123 cat_loss: 1.4184 data: 0.0302 batch: 0.0721 eta: 0:16:35
2023-04-18 11:33:51,537 - lanedet.utils.recorder - INFO - epoch: 13 step: 10953 lr: 0.0145 det_loss: 0.5849 cat_loss: 1.4232 data: 0.0338 batch: 0.0878 eta: 0:16:28
2023-04-18 11:34:24,622 - lanedet.utils.recorder - INFO - Best detection metric: 0.9217116724738679 Best classification metric: 0.7463592233009712
2023-04-18 11:34:25,840 - lanedet.utils.recorder - INFO - epoch: 14 step: 11257 lr: 0.0142 det_loss: 0.5693 cat_loss: 1.4208 data: 0.0804 batch: 0.1213 eta: 0:16:06
2023-04-18 11:34:33,459 - lanedet.utils.recorder - INFO - epoch: 14 step: 11357 lr: 0.0141 det_loss: 0.5667 cat_loss: 1.4291 data: 0.0293 batch: 0.0720 eta: 0:15:59
2023-04-18 11:34:42,462 - lanedet.utils.recorder - INFO - epoch: 14 step: 11457 lr: 0.0140 det_loss: 0.5431 cat_loss: 1.3885 data: 0.0367 batch: 0.0870 eta: 0:15:53
2023-04-18 11:34:50,487 - lanedet.utils.recorder - INFO - epoch: 14 step: 11557 lr: 0.0139 det_loss: 0.6052 cat_loss: 1.4469 data: 0.0293 batch: 0.0760 eta: 0:15:46
2023-04-18 11:34:58,318 - lanedet.utils.recorder - INFO - epoch: 14 step: 11657 lr: 0.0138 det_loss: 0.5528 cat_loss: 1.4199 data: 0.0306 batch: 0.0768 eta: 0:15:39
2023-04-18 11:35:06,700 - lanedet.utils.recorder - INFO - epoch: 14 step: 11757 lr: 0.0137 det_loss: 0.5996 cat_loss: 1.4357 data: 0.0304 batch: 0.0833 eta: 0:15:32
2023-04-18 11:35:40,826 - lanedet.utils.recorder - INFO - Detection: 0.9268546747967479 classification accuracy: 0.7257281553398057
2023-04-18 11:35:45,011 - lanedet.utils.recorder - INFO - Best detection metric: 0.9268546747967479 Best classification metric: 0.7463592233009712
2023-04-18 11:35:46,326 - lanedet.utils.recorder - INFO - epoch: 15 step: 12061 lr: 0.0134 det_loss: 0.5159 cat_loss: 1.4343 data: 0.0849 batch: 0.1282 eta: 0:15:12
2023-04-18 11:35:54,383 - lanedet.utils.recorder - INFO - epoch: 15 step: 12161 lr: 0.0133 det_loss: 0.5514 cat_loss: 1.4319 data: 0.0293 batch: 0.0737 eta: 0:15:05
2023-04-18 11:36:01,919 - lanedet.utils.recorder - INFO - epoch: 15 step: 12261 lr: 0.0132 det_loss: 0.5877 cat_loss: 1.4399 data: 0.0295 batch: 0.0757 eta: 0:14:57
2023-04-18 11:36:09,847 - lanedet.utils.recorder - INFO - epoch: 15 step: 12361 lr: 0.0131 det_loss: 0.5653 cat_loss: 1.4393 data: 0.0317 batch: 0.0738 eta: 0:14:50
2023-04-18 11:36:17,608 - lanedet.utils.recorder - INFO - epoch: 15 step: 12461 lr: 0.0130 det_loss: 0.5644 cat_loss: 1.3930 data: 0.0321 batch: 0.0767 eta: 0:14:43
2023-04-18 11:36:24,579 - lanedet.utils.recorder - INFO - epoch: 15 step: 12561 lr: 0.0129 det_loss: 0.5434 cat_loss: 1.3969 data: 0.0294 batch: 0.0698 eta: 0:14:35
2023-04-18 11:36:32,792 - lanedet.utils.recorder - INFO - epoch: 15 step: 12661 lr: 0.0128 det_loss: 0.5062 cat_loss: 1.3934 data: 0.0366 batch: 0.0928 eta: 0:14:28
2023-04-18 11:36:58,539 - lanedet.utils.recorder - INFO - Detection: 0.917149390243903 classification accuracy: 0.7508090614886728
2023-04-18 11:36:58,540 - lanedet.utils.recorder - INFO - Best detection metric: 0.9268546747967479 Best classification metric: 0.7508090614886728
2023-04-18 11:36:59,538 - lanedet.utils.recorder - INFO - epoch: 16 step: 12865 lr: 0.0126 det_loss: 0.5385 cat_loss: 1.4601 data: 0.0695 batch: 0.1089 eta: 0:14:13
2023-04-18 11:37:07,314 - lanedet.utils.recorder - INFO - epoch: 16 step: 12965 lr: 0.0125 det_loss: 0.5085 cat_loss: 1.4059 data: 0.0321 batch: 0.0745 eta: 0:14:05
2023-04-18 11:37:14,619 - lanedet.utils.recorder - INFO - epoch: 16 step: 13065 lr: 0.0124 det_loss: 0.5469 cat_loss: 1.4058 data: 0.0279 batch: 0.0713 eta: 0:13:58
2023-04-18 11:37:21,736 - lanedet.utils.recorder - INFO - epoch: 16 step: 13165 lr: 0.0123 det_loss: 0.5553 cat_loss: 1.4365 data: 0.0279 batch: 0.0717 eta: 0:13:50
2023-04-18 11:37:28,735 - lanedet.utils.recorder - INFO - epoch: 16 step: 13265 lr: 0.0122 det_loss: 0.5658 cat_loss: 1.4275 data: 0.0295 batch: 0.0705 eta: 0:13:42
2023-04-18 11:37:35,874 - lanedet.utils.recorder - INFO - epoch: 16 step: 13365 lr: 0.0121 det_loss: 0.5435 cat_loss: 1.4073 data: 0.0306 batch: 0.0722 eta: 0:13:34
2023-04-18 11:37:43,113 - lanedet.utils.recorder - INFO - epoch: 16 step: 13465 lr: 0.0120 det_loss: 0.4869 cat_loss: 1.4348 data: 0.0283 batch: 0.0750 eta: 0:13:26
2023-04-18 11:37:50,586 - lanedet.utils.recorder - INFO - epoch: 16 step: 13565 lr: 0.0119 det_loss: 0.5318 cat_loss: 1.3971 data: 0.0280 batch: 0.0734 eta: 0:13:18
2023-04-18 11:37:57,578 - lanedet.utils.recorder - INFO - epoch: 16 step: 13665 lr: 0.0118 det_loss: 0.5818 cat_loss: 1.4634 data: 0.0287 batch: 0.0672 eta: 0:13:10
2023-04-18 11:37:57,772 - lanedet.utils.recorder - INFO - epoch: 16 step: 13668 lr: 0.0118 det_loss: 0.5797 cat_loss: 1.4549 data: 0.0292 batch: 0.0670 eta: 0:13:10
2023-04-18 11:38:08,032 - lanedet.utils.recorder - INFO - Detection: 0.9268111207897795 classification accuracy: 0.6759708737864079
2023-04-18 11:38:08,032 - lanedet.utils.recorder - INFO - Best detection metric: 0.9268546747967479 Best classification metric: 0.7508090614886728
2023-04-18 11:38:09,179 - lanedet.utils.recorder - INFO - epoch: 17 step: 13669 lr: 0.0118 det_loss: 0.5791 cat_loss: 1.4549 data: 0.0800 batch: 0.1184 eta: 0:13:11
2023-04-18 11:38:16,469 - lanedet.utils.recorder - INFO - epoch: 17 step: 13769 lr: 0.0117 det_loss: 0.5340 cat_loss: 1.4260 data: 0.0283 batch: 0.0692 eta: 0:13:03
2023-04-18 11:38:23,671 - lanedet.utils.recorder - INFO - epoch: 17 step: 13869 lr: 0.0116 det_loss: 0.5553 cat_loss: 1.3909 data: 0.0287 batch: 0.0717 eta: 0:12:55
2023-04-18 11:38:30,841 - lanedet.utils.recorder - INFO - epoch: 17 step: 13969 lr: 0.0115 det_loss: 0.5521 cat_loss: 1.4158 data: 0.0273 batch: 0.0728 eta: 0:12:48
2023-04-18 11:38:38,003 - lanedet.utils.recorder - INFO - epoch: 17 step: 14069 lr: 0.0114 det_loss: 0.5446 cat_loss: 1.4215 data: 0.0302 batch: 0.0710 eta: 0:12:40
2023-04-18 11:38:45,152 - lanedet.utils.recorder - INFO - epoch: 17 step: 14169 lr: 0.0113 det_loss: 0.5502 cat_loss: 1.4219 data: 0.0297 batch: 0.0721 eta: 0:12:32
2023-04-18 11:38:52,173 - lanedet.utils.recorder - INFO - epoch: 17 step: 14269 lr: 0.0112 det_loss: 0.5025 cat_loss: 1.4323 data: 0.0274 batch: 0.0693 eta: 0:12:24
2023-04-18 11:39:17,124 - lanedet.utils.recorder - INFO - Detection: 0.9215737514517993 classification accuracy: 0.7637540453074433
2023-04-18 11:39:17,124 - lanedet.utils.recorder - INFO - Best detection metric: 0.9268546747967479 Best classification metric: 0.7637540453074433
2023-04-18 11:39:18,182 - lanedet.utils.recorder - INFO - epoch: 18 step: 14473 lr: 0.0110 det_loss: 0.5010 cat_loss: 1.3739 data: 0.0739 batch: 0.1146 eta: 0:12:09
2023-04-18 11:39:25,899 - lanedet.utils.recorder - INFO - epoch: 18 step: 14573 lr: 0.0109 det_loss: 0.5126 cat_loss: 1.3879 data: 0.0281 batch: 0.0722 eta: 0:12:01
2023-04-18 11:39:33,384 - lanedet.utils.recorder - INFO - epoch: 18 step: 14673 lr: 0.0108 det_loss: 0.5243 cat_loss: 1.4218 data: 0.0276 batch: 0.0723 eta: 0:11:54
2023-04-18 11:39:40,698 - lanedet.utils.recorder - INFO - epoch: 18 step: 14773 lr: 0.0107 det_loss: 0.4700 cat_loss: 1.4006 data: 0.0287 batch: 0.0736 eta: 0:11:46
2023-04-18 11:39:47,937 - lanedet.utils.recorder - INFO - epoch: 18 step: 14873 lr: 0.0106 det_loss: 0.5152 cat_loss: 1.4023 data: 0.0306 batch: 0.0748 eta: 0:11:38
2023-04-18 11:39:55,316 - lanedet.utils.recorder - INFO - epoch: 18 step: 14973 lr: 0.0105 det_loss: 0.5356 cat_loss: 1.4177 data: 0.0285 batch: 0.0733 eta: 0:11:31
2023-04-18 11:40:27,818 - lanedet.utils.recorder - INFO - Detection: 0.9308471254355404 classification accuracy: 0.7544498381877021
2023-04-18 11:40:31,042 - lanedet.utils.recorder - INFO - Best detection metric: 0.9308471254355404 Best classification metric: 0.7637540453074433
2023-04-18 11:40:32,156 - lanedet.utils.recorder - INFO - epoch: 19 step: 15277 lr: 0.0102 det_loss: 0.5017 cat_loss: 1.4279 data: 0.0767 batch: 0.1183 eta: 0:11:08
2023-04-18 11:40:39,720 - lanedet.utils.recorder - INFO - epoch: 19 step: 15377 lr: 0.0100 det_loss: 0.4798 cat_loss: 1.4199 data: 0.0273 batch: 0.0714 eta: 0:11:00
2023-04-18 11:40:47,189 - lanedet.utils.recorder - INFO - epoch: 19 step: 15477 lr: 0.0099 det_loss: 0.5343 cat_loss: 1.4220 data: 0.0272 batch: 0.0702 eta: 0:10:53
2023-04-18 11:40:55,037 - lanedet.utils.recorder - INFO - epoch: 19 step: 15577 lr: 0.0098 det_loss: 0.5089 cat_loss: 1.3726 data: 0.0303 batch: 0.0728 eta: 0:10:45
2023-04-18 11:41:02,533 - lanedet.utils.recorder - INFO - epoch: 19 step: 15677 lr: 0.0097 det_loss: 0.5185 cat_loss: 1.3924 data: 0.0285 batch: 0.0703 eta: 0:10:38
2023-04-18 11:41:10,020 - lanedet.utils.recorder - INFO - epoch: 19 step: 15777 lr: 0.0096 det_loss: 0.5155 cat_loss: 1.4094 data: 0.0306 batch: 0.0772 eta: 0:10:30
2023-04-18 11:41:42,546 - lanedet.utils.recorder - INFO - Detection: 0.930730981416957 classification accuracy: 0.7637540453074434
2023-04-18 11:41:42,546 - lanedet.utils.recorder - INFO - Best detection metric: 0.9308471254355404 Best classification metric: 0.7637540453074434
2023-04-18 11:41:43,710 - lanedet.utils.recorder - INFO - epoch: 20 step: 16081 lr: 0.0093 det_loss: 0.4823 cat_loss: 1.4049 data: 0.0787 batch: 0.1198 eta: 0:10:08
2023-04-18 11:41:51,135 - lanedet.utils.recorder - INFO - epoch: 20 step: 16181 lr: 0.0092 det_loss: 0.4930 cat_loss: 1.3649 data: 0.0305 batch: 0.0707 eta: 0:10:00
2023-04-18 11:41:58,350 - lanedet.utils.recorder - INFO - epoch: 20 step: 16281 lr: 0.0091 det_loss: 0.4993 cat_loss: 1.4094 data: 0.0280 batch: 0.0701 eta: 0:09:52
2023-04-18 11:42:05,714 - lanedet.utils.recorder - INFO - epoch: 20 step: 16381 lr: 0.0090 det_loss: 0.4871 cat_loss: 1.4220 data: 0.0283 batch: 0.0739 eta: 0:09:45
2023-04-18 11:42:13,827 - lanedet.utils.recorder - INFO - epoch: 20 step: 16481 lr: 0.0089 det_loss: 0.5218 cat_loss: 1.4053 data: 0.0319 batch: 0.0790 eta: 0:09:37
2023-04-18 11:42:21,237 - lanedet.utils.recorder - INFO - epoch: 20 step: 16581 lr: 0.0088 det_loss: 0.4788 cat_loss: 1.4137 data: 0.0287 batch: 0.0742 eta: 0:09:30
2023-04-18 11:42:53,732 - lanedet.utils.recorder - INFO - Detection: 0.9323315911730545 classification accuracy: 0.7839805825242713
2023-04-18 11:42:57,975 - lanedet.utils.recorder - INFO - Best detection metric: 0.9323315911730545 Best classification metric: 0.7839805825242713
2023-04-18 11:42:59,019 - lanedet.utils.recorder - INFO - epoch: 21 step: 16885 lr: 0.0085 det_loss: 0.4869 cat_loss: 1.4116 data: 0.0740 batch: 0.1137 eta: 0:09:07
2023-04-18 11:43:06,430 - lanedet.utils.recorder - INFO - epoch: 21 step: 16985 lr: 0.0084 det_loss: 0.4595 cat_loss: 1.4322 data: 0.0275 batch: 0.0712 eta: 0:09:00
2023-04-18 11:43:13,648 - lanedet.utils.recorder - INFO - epoch: 21 step: 17085 lr: 0.0083 det_loss: 0.5099 cat_loss: 1.4181 data: 0.0269 batch: 0.0713 eta: 0:08:52
2023-04-18 11:43:20,863 - lanedet.utils.recorder - INFO - epoch: 21 step: 17185 lr: 0.0082 det_loss: 0.4936 cat_loss: 1.3990 data: 0.0267 batch: 0.0716 eta: 0:08:44
2023-04-18 11:43:27,948 - lanedet.utils.recorder - INFO - epoch: 21 step: 17285 lr: 0.0081 det_loss: 0.4980 cat_loss: 1.3822 data: 0.0282 batch: 0.0712 eta: 0:08:36
2023-04-18 11:43:35,083 - lanedet.utils.recorder - INFO - epoch: 21 step: 17385 lr: 0.0080 det_loss: 0.4600 cat_loss: 1.3719 data: 0.0282 batch: 0.0722 eta: 0:08:29
2023-04-18 11:43:42,328 - lanedet.utils.recorder - INFO - epoch: 21 step: 17485 lr: 0.0078 det_loss: 0.4974 cat_loss: 1.4011 data: 0.0276 batch: 0.0710 eta: 0:08:21
2023-04-18 11:44:08,227 - lanedet.utils.recorder - INFO - Detection: 0.9276204994192799 classification accuracy: 0.7661812297734628
2023-04-18 11:44:08,227 - lanedet.utils.recorder - INFO - Best detection metric: 0.9323315911730545 Best classification metric: 0.7839805825242713
2023-04-18 11:44:09,166 - lanedet.utils.recorder - INFO - epoch: 22 step: 17689 lr: 0.0076 det_loss: 0.4974 cat_loss: 1.4130 data: 0.0683 batch: 0.1081 eta: 0:08:06
2023-04-18 11:44:16,986 - lanedet.utils.recorder - INFO - epoch: 22 step: 17789 lr: 0.0075 det_loss: 0.4522 cat_loss: 1.3719 data: 0.0271 batch: 0.0715 eta: 0:07:58
2023-04-18 11:44:24,334 - lanedet.utils.recorder - INFO - epoch: 22 step: 17889 lr: 0.0074 det_loss: 0.4804 cat_loss: 1.3907 data: 0.0259 batch: 0.0722 eta: 0:07:51
2023-04-18 11:44:31,589 - lanedet.utils.recorder - INFO - epoch: 22 step: 17989 lr: 0.0073 det_loss: 0.4419 cat_loss: 1.3897 data: 0.0290 batch: 0.0717 eta: 0:07:43
2023-04-18 11:44:38,985 - lanedet.utils.recorder - INFO - epoch: 22 step: 18089 lr: 0.0072 det_loss: 0.4434 cat_loss: 1.4034 data: 0.0300 batch: 0.0733 eta: 0:07:36
2023-04-18 11:44:46,230 - lanedet.utils.recorder - INFO - epoch: 22 step: 18189 lr: 0.0071 det_loss: 0.4224 cat_loss: 1.3719 data: 0.0318 batch: 0.0784 eta: 0:07:28
2023-04-18 11:45:19,029 - lanedet.utils.recorder - INFO - Detection: 0.9373221544715442 classification accuracy: 0.7791262135922329
2023-04-18 11:45:23,100 - lanedet.utils.recorder - INFO - Best detection metric: 0.9373221544715442 Best classification metric: 0.7839805825242713
2023-04-18 11:45:24,142 - lanedet.utils.recorder - INFO - epoch: 23 step: 18493 lr: 0.0068 det_loss: 0.4221 cat_loss: 1.3867 data: 0.0713 batch: 0.1129 eta: 0:07:05
2023-04-18 11:45:31,561 - lanedet.utils.recorder - INFO - epoch: 23 step: 18593 lr: 0.0067 det_loss: 0.4381 cat_loss: 1.3969 data: 0.0298 batch: 0.0719 eta: 0:06:58
2023-04-18 11:45:38,844 - lanedet.utils.recorder - INFO - epoch: 23 step: 18693 lr: 0.0066 det_loss: 0.4735 cat_loss: 1.3817 data: 0.0296 batch: 0.0715 eta: 0:06:50
2023-04-18 11:45:46,487 - lanedet.utils.recorder - INFO - epoch: 23 step: 18793 lr: 0.0064 det_loss: 0.4676 cat_loss: 1.3832 data: 0.0309 batch: 0.0732 eta: 0:06:43
2023-04-18 11:45:54,154 - lanedet.utils.recorder - INFO - epoch: 23 step: 18893 lr: 0.0063 det_loss: 0.4218 cat_loss: 1.3858 data: 0.0316 batch: 0.0793 eta: 0:06:35
2023-04-18 11:46:01,501 - lanedet.utils.recorder - INFO - epoch: 23 step: 18993 lr: 0.0062 det_loss: 0.4343 cat_loss: 1.4069 data: 0.0299 batch: 0.0712 eta: 0:06:28
2023-04-18 11:46:09,075 - lanedet.utils.recorder - INFO - epoch: 23 step: 19093 lr: 0.0061 det_loss: 0.4326 cat_loss: 1.4303 data: 0.0297 batch: 0.0742 eta: 0:06:20
2023-04-18 11:46:35,003 - lanedet.utils.recorder - INFO - Detection: 0.929990563298491 classification accuracy: 0.7657766990291265
2023-04-18 11:46:35,004 - lanedet.utils.recorder - INFO - Best detection metric: 0.9373221544715442 Best classification metric: 0.7839805825242713
2023-04-18 11:46:36,046 - lanedet.utils.recorder - INFO - epoch: 24 step: 19297 lr: 0.0059 det_loss: 0.4546 cat_loss: 1.3965 data: 0.0707 batch: 0.1142 eta: 0:06:05
2023-04-18 11:46:43,898 - lanedet.utils.recorder - INFO - epoch: 24 step: 19397 lr: 0.0058 det_loss: 0.4502 cat_loss: 1.3925 data: 0.0334 batch: 0.0772 eta: 0:05:58
2023-04-18 11:46:51,761 - lanedet.utils.recorder - INFO - epoch: 24 step: 19497 lr: 0.0057 det_loss: 0.4399 cat_loss: 1.3887 data: 0.0306 batch: 0.0766 eta: 0:05:50
2023-04-18 11:46:59,126 - lanedet.utils.recorder - INFO - epoch: 24 step: 19597 lr: 0.0056 det_loss: 0.4867 cat_loss: 1.4003 data: 0.0300 batch: 0.0737 eta: 0:05:42
2023-04-18 11:47:06,858 - lanedet.utils.recorder - INFO - epoch: 24 step: 19697 lr: 0.0055 det_loss: 0.4650 cat_loss: 1.4157 data: 0.0302 batch: 0.0750 eta: 0:05:35
2023-04-18 11:47:14,568 - lanedet.utils.recorder - INFO - epoch: 24 step: 19797 lr: 0.0053 det_loss: 0.4477 cat_loss: 1.3809 data: 0.0311 batch: 0.0864 eta: 0:05:27
2023-04-18 11:47:22,004 - lanedet.utils.recorder - INFO - epoch: 24 step: 19897 lr: 0.0052 det_loss: 0.4415 cat_loss: 1.4048 data: 0.0309 batch: 0.0759 eta: 0:05:20
2023-04-18 11:47:48,053 - lanedet.utils.recorder - INFO - Detection: 0.9381133855981414 classification accuracy: 0.7815533980582525
2023-04-18 11:47:52,149 - lanedet.utils.recorder - INFO - Best detection metric: 0.9381133855981414 Best classification metric: 0.7839805825242713
2023-04-18 11:47:53,202 - lanedet.utils.recorder - INFO - epoch: 25 step: 20101 lr: 0.0050 det_loss: 0.4367 cat_loss: 1.4197 data: 0.0722 batch: 0.1148 eta: 0:05:05
2023-04-18 11:48:01,196 - lanedet.utils.recorder - INFO - epoch: 25 step: 20201 lr: 0.0049 det_loss: 0.4675 cat_loss: 1.3990 data: 0.0276 batch: 0.0691 eta: 0:04:57
2023-04-18 11:48:08,718 - lanedet.utils.recorder - INFO - epoch: 25 step: 20301 lr: 0.0048 det_loss: 0.4272 cat_loss: 1.4119 data: 0.0322 batch: 0.0819 eta: 0:04:50
2023-04-18 11:48:16,101 - lanedet.utils.recorder - INFO - epoch: 25 step: 20401 lr: 0.0047 det_loss: 0.4557 cat_loss: 1.3944 data: 0.0269 batch: 0.0691 eta: 0:04:42
2023-04-18 11:48:23,749 - lanedet.utils.recorder - INFO - epoch: 25 step: 20501 lr: 0.0046 det_loss: 0.4084 cat_loss: 1.3760 data: 0.0311 batch: 0.0845 eta: 0:04:35
2023-04-18 11:48:31,637 - lanedet.utils.recorder - INFO - epoch: 25 step: 20601 lr: 0.0045 det_loss: 0.4639 cat_loss: 1.3774 data: 0.0293 batch: 0.0861 eta: 0:04:27
2023-04-18 11:49:05,599 - lanedet.utils.recorder - INFO - Detection: 0.9368394308943088 classification accuracy: 0.7807443365695793
2023-04-18 11:49:05,599 - lanedet.utils.recorder - INFO - Best detection metric: 0.9381133855981414 Best classification metric: 0.7839805825242713
2023-04-18 11:49:06,886 - lanedet.utils.recorder - INFO - epoch: 26 step: 20905 lr: 0.0041 det_loss: 0.4256 cat_loss: 1.4017 data: 0.0831 batch: 0.1252 eta: 0:04:05
2023-04-18 11:49:14,268 - lanedet.utils.recorder - INFO - epoch: 26 step: 21005 lr: 0.0040 det_loss: 0.4540 cat_loss: 1.3823 data: 0.0297 batch: 0.0695 eta: 0:03:57
2023-04-18 11:49:21,609 - lanedet.utils.recorder - INFO - epoch: 26 step: 21105 lr: 0.0039 det_loss: 0.4311 cat_loss: 1.3996 data: 0.0313 batch: 0.0702 eta: 0:03:49
2023-04-18 11:49:29,354 - lanedet.utils.recorder - INFO - epoch: 26 step: 21205 lr: 0.0038 det_loss: 0.4150 cat_loss: 1.4074 data: 0.0337 batch: 0.0790 eta: 0:03:42
2023-04-18 11:49:37,528 - lanedet.utils.recorder - INFO - epoch: 26 step: 21305 lr: 0.0036 det_loss: 0.4322 cat_loss: 1.3782 data: 0.0332 batch: 0.0762 eta: 0:03:34
2023-04-18 11:49:44,981 - lanedet.utils.recorder - INFO - epoch: 26 step: 21405 lr: 0.0035 det_loss: 0.4116 cat_loss: 1.3594 data: 0.0281 batch: 0.0752 eta: 0:03:27
2023-04-18 11:49:52,673 - lanedet.utils.recorder - INFO - epoch: 26 step: 21505 lr: 0.0034 det_loss: 0.4281 cat_loss: 1.3823 data: 0.0335 batch: 0.0822 eta: 0:03:19
2023-04-18 11:50:18,524 - lanedet.utils.recorder - INFO - Detection: 0.9392385307781648 classification accuracy: 0.7815533980582523
2023-04-18 11:50:22,666 - lanedet.utils.recorder - INFO - Best detection metric: 0.9392385307781648 Best classification metric: 0.7839805825242713
2023-04-18 11:50:23,791 - lanedet.utils.recorder - INFO - epoch: 27 step: 21709 lr: 0.0032 det_loss: 0.4164 cat_loss: 1.4111 data: 0.0767 batch: 0.1173 eta: 0:03:04
2023-04-18 11:50:31,331 - lanedet.utils.recorder - INFO - epoch: 27 step: 21809 lr: 0.0031 det_loss: 0.4191 cat_loss: 1.3990 data: 0.0297 batch: 0.0707 eta: 0:02:56
2023-04-18 11:50:38,900 - lanedet.utils.recorder - INFO - epoch: 27 step: 21909 lr: 0.0029 det_loss: 0.4184 cat_loss: 1.4031 data: 0.0300 batch: 0.0679 eta: 0:02:49
2023-04-18 11:50:45,950 - lanedet.utils.recorder - INFO - epoch: 27 step: 22009 lr: 0.0028 det_loss: 0.4584 cat_loss: 1.3657 data: 0.0326 batch: 0.0753 eta: 0:02:41
2023-04-18 11:50:53,480 - lanedet.utils.recorder - INFO - epoch: 27 step: 22109 lr: 0.0027 det_loss: 0.4163 cat_loss: 1.3718 data: 0.0275 batch: 0.0716 eta: 0:02:34
2023-04-18 11:51:00,649 - lanedet.utils.recorder - INFO - epoch: 27 step: 22209 lr: 0.0026 det_loss: 0.4165 cat_loss: 1.3740 data: 0.0279 batch: 0.0716 eta: 0:02:26
2023-04-18 11:51:07,686 - lanedet.utils.recorder - INFO - epoch: 27 step: 22309 lr: 0.0025 det_loss: 0.4092 cat_loss: 1.4031 data: 0.0283 batch: 0.0714 eta: 0:02:19
2023-04-18 11:51:33,141 - lanedet.utils.recorder - INFO - Detection: 0.9382730836236924 classification accuracy: 0.7771035598705499
2023-04-18 11:51:33,141 - lanedet.utils.recorder - INFO - Best detection metric: 0.9392385307781648 Best classification metric: 0.7839805825242713
2023-04-18 11:51:34,343 - lanedet.utils.recorder - INFO - epoch: 28 step: 22513 lr: 0.0022 det_loss: 0.4327 cat_loss: 1.3928 data: 0.0809 batch: 0.1231 eta: 0:02:03
2023-04-18 11:51:41,741 - lanedet.utils.recorder - INFO - epoch: 28 step: 22613 lr: 0.0021 det_loss: 0.4171 cat_loss: 1.3873 data: 0.0265 batch: 0.0720 eta: 0:01:56
2023-04-18 11:51:48,961 - lanedet.utils.recorder - INFO - epoch: 28 step: 22713 lr: 0.0020 det_loss: 0.4192 cat_loss: 1.3963 data: 0.0265 batch: 0.0710 eta: 0:01:48
2023-04-18 11:51:56,338 - lanedet.utils.recorder - INFO - epoch: 28 step: 22813 lr: 0.0018 det_loss: 0.4211 cat_loss: 1.3945 data: 0.0284 batch: 0.0708 eta: 0:01:40
2023-04-18 11:52:03,481 - lanedet.utils.recorder - INFO - epoch: 28 step: 22913 lr: 0.0017 det_loss: 0.4208 cat_loss: 1.3511 data: 0.0282 batch: 0.0702 eta: 0:01:33
2023-04-18 11:52:10,614 - lanedet.utils.recorder - INFO - epoch: 28 step: 23013 lr: 0.0016 det_loss: 0.3902 cat_loss: 1.3948 data: 0.0269 batch: 0.0713 eta: 0:01:25
2023-04-18 11:52:17,713 - lanedet.utils.recorder - INFO - epoch: 28 step: 23113 lr: 0.0015 det_loss: 0.3755 cat_loss: 1.3824 data: 0.0283 batch: 0.0718 eta: 0:01:18
2023-04-18 11:52:42,452 - lanedet.utils.recorder - INFO - Detection: 0.9387158826945403 classification accuracy: 0.792071197411003
2023-04-18 11:52:42,452 - lanedet.utils.recorder - INFO - Best detection metric: 0.9392385307781648 Best classification metric: 0.792071197411003
2023-04-18 11:52:43,391 - lanedet.utils.recorder - INFO - epoch: 29 step: 23317 lr: 0.0012 det_loss: 0.3697 cat_loss: 1.3406 data: 0.0659 batch: 0.1072 eta: 0:01:02
2023-04-18 11:52:50,891 - lanedet.utils.recorder - INFO - epoch: 29 step: 23417 lr: 0.0011 det_loss: 0.3762 cat_loss: 1.3761 data: 0.0301 batch: 0.0711 eta: 0:00:55
2023-04-18 11:52:58,044 - lanedet.utils.recorder - INFO - epoch: 29 step: 23517 lr: 0.0009 det_loss: 0.4473 cat_loss: 1.3907 data: 0.0289 batch: 0.0710 eta: 0:00:47
2023-04-18 11:53:05,091 - lanedet.utils.recorder - INFO - epoch: 29 step: 23617 lr: 0.0008 det_loss: 0.4759 cat_loss: 1.4219 data: 0.0301 batch: 0.0701 eta: 0:00:40
2023-04-18 11:53:12,238 - lanedet.utils.recorder - INFO - epoch: 29 step: 23717 lr: 0.0007 det_loss: 0.3696 cat_loss: 1.3660 data: 0.0266 batch: 0.0706 eta: 0:00:32
2023-04-18 11:53:19,721 - lanedet.utils.recorder - INFO - epoch: 29 step: 23817 lr: 0.0005 det_loss: 0.4078 cat_loss: 1.3990 data: 0.0303 batch: 0.0817 eta: 0:00:25
2023-04-18 11:53:26,860 - lanedet.utils.recorder - INFO - epoch: 29 step: 23917 lr: 0.0004 det_loss: 0.3779 cat_loss: 1.3843 data: 0.0305 batch: 0.0712 eta: 0:00:17
2023-04-18 11:53:57,028 - lanedet.utils.recorder - INFO - Detection: 0.9408209930313587 classification accuracy: 0.7896440129449837
2023-04-18 11:54:01,258 - lanedet.utils.recorder - INFO - Best detection metric: 0.9408209930313587 Best classification metric: 0.792071197411003
This is the loss function:
def loss(self, output, batch):
criterion = SoftmaxFocalLoss(2)
total_loss = 0
loss_stats = {}
det_loss = criterion(output['cls'], batch['cls_label'])
loss_fn = torch.nn.CrossEntropyLoss()
score = F.softmax(output['category'], dim=1)
cat_loss = loss_fn(score, batch['category'])
loss_stats.update({'det_loss': det_loss, 'cat_loss': cat_loss})
total_loss = det_loss + cat_loss
ret = {'loss': total_loss , 'loss_stats': loss_stats}
return ret
@zillur-av
I think there might be two problems:
- remove the line of
score = F.softmax(output['category'], dim=1) - check if the order of lane categories correspond to the order of lanes
@zillur-av I think there might be two problems:
- remove the line of
score = F.softmax(output['category'], dim=1)- check if the order of lane categories correspond to the order of lanes
- IF I remove
softmaxfunction, the summation of all the class probabilities of each label will not be equal to 1. Then how can I compute loss? - I believe order is ok.
@zillur-av
If I'm correct, torch.nn.CrossEntropyLoss handles the input without softmax. You only need softmax when you are in post-processing, like something when you want to get the probablity of each category during inference.
@zillur-av How about the classification accuracy?
@zillur-av How about the classification accuracy?
Please check this repo https://github.com/zillur-av/LVLane