I’m running the code on docker and the training happens without giving any error, but as you know its difficult to view the metrics on docker I wanted to know where do you configure the tensor board url and port number in the project
Also the accuracy is low, the change I’ve made in the code is setting learning rate to 0.0001
Technically setting lower learning rate help the model in finding better local minima but how does 0.004 help in your case??
During evaluation, there are two different metrics that are getting printed does it correspond to the teacher and student model?
Metrics obtained:
[07/01 02:50:11 d2.evaluation.fast_eval_api]: Evaluate annotation type bbox
[07/01 02:50:11 d2.evaluation.fast_eval_api]: COCOeval_opt.evaluate() finished in 0.00 seconds.
[07/01 02:50:11 d2.evaluation.fast_eval_api]: Accumulating evaluation results..
[07/01 02:50:11 d2.evaluation.fast_eval_api]: COCOeval_opt.accumulate() finished in 0.01 seconds.
Average Precision(AP) @[ IoU=0.50:0.95 | area=all | maDets= 100 ] = 0.000
Average Precision(AP) @[ IoU=0.50area=all I maxDets=100 = 0.001
Average Precision(AP) @[ IoU=0.75area=all | maDets= 100 ] = 0.000
Average Precision(AP) @ IoU=0.50:0.95 | area= small | maDets= 100 ] = 0.000
Average Precision(AP) a[ IoU=0.50:0.95 | area=medium | maDets= 100 ] = 0.000
Average Precision(AP) a[ IoU=0.50:0.95 | area= large | maDets= 100 ] = 0.000
Average Recall(AR) @[ IOU=0.50:0.95area= all I maxDets= 1 1= 0.001
Average Recall(AR) a[ IOU=0.50:0.95area= all | maDets= 10 1= 0.001
Average Recall (AR) @[ IOU=0.50:0.95 area= all I maDets= 100 ] = 0.001
Average Recall(AR) @[ IoU=0.50:0.95 area= small | maxDets=100 ] = 0.000
Average Recall (AR) @[ IoU=0.50:0.95 | area-medium | maDets=100 ] = 0.001
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maDets= 100 ] = 0.000
[07/01 02:50:11 d2.engine.defaults]: Evaluation results for hit_test in cv format:
[07/01 02:50:11 d2.evaluation. testing]: copypaste: Task: box
[07/01 02:50:11 d2.evaluation. testing]: copypaste: AP, AP50, AP75, APs, APm, API [07/01 02:50:11 d2.evaluation.testing]: copypaste: 0.0297,0.0990,0.0000,0.0000,0.0297,0.0000
DONE (t=0.00s)
creating index...
index created!
[07/01 02:50:13 d2.evaluation.fast_eval_api]: Evaluate annotation type bbox
[07/01 02:50:13 d2.evaluation. fast_eval_api]: COCOevalopt.evaluate() finished in 0.01 seconds.
[07/01 02:50:13 d2.evaluation.fast eval_ apil: Accumulating evaluation results..
[07/01 02:50:13 d2.evaluation.fast_eval_api]: cocoeval_opt.accumulate() finished in 0.01 seconds.
Average Precision (AP) @[ IoU=0.50:0.95 | area= ali | maxDets=100 j = 0.001
Average Precision (AP) a[ IOU=0.50 area= all | maxDets= 100 0.004
Average Precision (AP) @[ IoU=0.75 area= all I maxDets=100 0.000
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maDets= 100 0.000
Average Precision (AP) al IOU=0.50:0.95 area=medium | maDets= 100 0.000
Average Precision (AP) a[ IoU=0.50:0.95 area= large |maxDets=100 0.005
Average Recall (AR) @[ IOU=0.50:0.95 area= all 0.001
Average Recall (AR) @[ IoU=0.50:0.95 | area= all maxDets= 10 0.005
Average Recall (AR) @[ IoU=0.50:0.95 | area= all maDets=100 = 0.006
Average Recall (AR) a[ IoU=0.50:0.95 | area= small maDets=100 = 0.000
Average Recall (AR) a[ IoU=0.50:0.95 | area=medium maDets=100 = 0.002
Average Recall (AR) a[ IoU=0.50:0.95 | area= large maDets=100 ] = 0.017
[07/01 02:50:13 d2.engine.defaults]: Evaluation results for hit_test in csv format:
[07/01 02:50:13 d2.evaluation.testing]: copypaste: Task: box
[07/01 02:50:13 d2.evaluation.testing]: copypaste: AP,AP50, AP75, APs, APm,APl [07/01 02:50:13 d2.evaluation.testing]: copypaste: 0.0684,0.4279,0.0261,0.0000,0.0418, 0.5054
[07/01 02:50:13 d2.utils.events]: eta: 7:46:02 iter: 71999 total_loss: 1.961 loss_cls: 0.1035 loss_box reg: 0.1161 loss_rpncls: 0.05318 loss rpn_loc: 0.02835 Loss_D_img_s: 0.6801 1 oss_cls_pseudo: 0.06015 loss_box_reg_pseudo: 0.1409 loss_rp_cis_pseudo: 0.02776 loss_rpn_loc_pseudo: 0.008429 loss_D_img_t: 0.6915 time: 0.9060 data_time: 0.0105 lr: 0.0001 max mem: 17580M
[07/01 02:50:33 d2.utils.events]: eta: 7:44:59 iter: 72019 total loss: 1.888 loss_cls: 0.09791 loss_box_reg: 0.1024 loss_rp_cls: 0.05297 loss_rpn_loc: 0.02758 loss_D_img_s: 0.6845 loss_cis_pseudo: 0.05373 loss_box_reg_pseudo: 0.1411 loss_rpn_cis_pseudo: 0.01927 loss_rpn_loc_pseudo: 0.00774 loss_D_img_t: 0.6896 time: 0.9061 17580M
I’m running the code on docker and the training happens without giving any error, but as you know its difficult to view the metrics on docker I wanted to know where do you configure the tensor board url and port number in the project
Also the accuracy is low, the change I’ve made in the code is setting learning rate to 0.0001 Technically setting lower learning rate help the model in finding better local minima but how does 0.004 help in your case??
During evaluation, there are two different metrics that are getting printed does it correspond to the teacher and student model?
Metrics obtained: [07/01 02:50:11 d2.evaluation.fast_eval_api]: Evaluate annotation type bbox [07/01 02:50:11 d2.evaluation.fast_eval_api]: COCOeval_opt.evaluate() finished in 0.00 seconds. [07/01 02:50:11 d2.evaluation.fast_eval_api]: Accumulating evaluation results.. [07/01 02:50:11 d2.evaluation.fast_eval_api]: COCOeval_opt.accumulate() finished in 0.01 seconds. Average Precision(AP) @[ IoU=0.50:0.95 | area=all | maDets= 100 ] = 0.000 Average Precision(AP) @[ IoU=0.50area=all I maxDets=100 = 0.001 Average Precision(AP) @[ IoU=0.75area=all | maDets= 100 ] = 0.000 Average Precision(AP) @ IoU=0.50:0.95 | area= small | maDets= 100 ] = 0.000 Average Precision(AP) a[ IoU=0.50:0.95 | area=medium | maDets= 100 ] = 0.000 Average Precision(AP) a[ IoU=0.50:0.95 | area= large | maDets= 100 ] = 0.000 Average Recall(AR) @[ IOU=0.50:0.95area= all I maxDets= 1 1= 0.001 Average Recall(AR) a[ IOU=0.50:0.95area= all | maDets= 10 1= 0.001 Average Recall (AR) @[ IOU=0.50:0.95 area= all I maDets= 100 ] = 0.001 Average Recall(AR) @[ IoU=0.50:0.95 area= small | maxDets=100 ] = 0.000 Average Recall (AR) @[ IoU=0.50:0.95 | area-medium | maDets=100 ] = 0.001 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maDets= 100 ] = 0.000 [07/01 02:50:11 d2.engine.defaults]: Evaluation results for hit_test in cv format: [07/01 02:50:11 d2.evaluation. testing]: copypaste: Task: box [07/01 02:50:11 d2.evaluation. testing]: copypaste: AP, AP50, AP75, APs, APm, API [07/01 02:50:11 d2.evaluation.testing]: copypaste: 0.0297,0.0990,0.0000,0.0000,0.0297,0.0000
DONE (t=0.00s) creating index... index created! [07/01 02:50:13 d2.evaluation.fast_eval_api]: Evaluate annotation type bbox [07/01 02:50:13 d2.evaluation. fast_eval_api]: COCOevalopt.evaluate() finished in 0.01 seconds. [07/01 02:50:13 d2.evaluation.fast eval_ apil: Accumulating evaluation results.. [07/01 02:50:13 d2.evaluation.fast_eval_api]: cocoeval_opt.accumulate() finished in 0.01 seconds. Average Precision (AP) @[ IoU=0.50:0.95 | area= ali | maxDets=100 j = 0.001 Average Precision (AP) a[ IOU=0.50 area= all | maxDets= 100 0.004 Average Precision (AP) @[ IoU=0.75 area= all I maxDets=100 0.000 Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maDets= 100 0.000 Average Precision (AP) al IOU=0.50:0.95 area=medium | maDets= 100 0.000 Average Precision (AP) a[ IoU=0.50:0.95 area= large |maxDets=100 0.005 Average Recall (AR) @[ IOU=0.50:0.95 area= all 0.001 Average Recall (AR) @[ IoU=0.50:0.95 | area= all maxDets= 10 0.005 Average Recall (AR) @[ IoU=0.50:0.95 | area= all maDets=100 = 0.006 Average Recall (AR) a[ IoU=0.50:0.95 | area= small maDets=100 = 0.000 Average Recall (AR) a[ IoU=0.50:0.95 | area=medium maDets=100 = 0.002 Average Recall (AR) a[ IoU=0.50:0.95 | area= large maDets=100 ] = 0.017 [07/01 02:50:13 d2.engine.defaults]: Evaluation results for hit_test in csv format: [07/01 02:50:13 d2.evaluation.testing]: copypaste: Task: box [07/01 02:50:13 d2.evaluation.testing]: copypaste: AP,AP50, AP75, APs, APm,APl [07/01 02:50:13 d2.evaluation.testing]: copypaste: 0.0684,0.4279,0.0261,0.0000,0.0418, 0.5054 [07/01 02:50:13 d2.utils.events]: eta: 7:46:02 iter: 71999 total_loss: 1.961 loss_cls: 0.1035 loss_box reg: 0.1161 loss_rpncls: 0.05318 loss rpn_loc: 0.02835 Loss_D_img_s: 0.6801 1 oss_cls_pseudo: 0.06015 loss_box_reg_pseudo: 0.1409 loss_rp_cis_pseudo: 0.02776 loss_rpn_loc_pseudo: 0.008429 loss_D_img_t: 0.6915 time: 0.9060 data_time: 0.0105 lr: 0.0001 max mem: 17580M [07/01 02:50:33 d2.utils.events]: eta: 7:44:59 iter: 72019 total loss: 1.888 loss_cls: 0.09791 loss_box_reg: 0.1024 loss_rp_cls: 0.05297 loss_rpn_loc: 0.02758 loss_D_img_s: 0.6845 loss_cis_pseudo: 0.05373 loss_box_reg_pseudo: 0.1411 loss_rpn_cis_pseudo: 0.01927 loss_rpn_loc_pseudo: 0.00774 loss_D_img_t: 0.6896 time: 0.9061 17580M