The results of validating a model which training with tridentnet_r101v2c4_c5_multiscale_addminival_3x_fp16.py are bad.

[zhayanli]

Describe the bug
Thanks for your excellent work！
I trained my own dataset using tridentnet. I converted my dataset to coco and ran create_coco_roidb.py , then changed gpus, num_class,log_frequency to 50,loader_worker to 4 in tridentnet_r101v2c4_c5_multiscale_addminival_3x_fp16.py. Do I need to change another parameters? I trained 10 epochs and used detection_test.py to validate. The results are bad.

Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.000
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.001
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.000
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.000
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.001
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.028
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.075
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.087
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.033
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.072
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.092

    I recorded RpnL1,RcnnL1 and Lr during the training. Could you help me, thank you!

Software info
driver, CUDA, cuDNN versions
OS verison
My Software info is Ubantu 16.04,CUDA 10.0,cuDNN 7.4.2

How did you set up your MXNet for SimpleDet

Additional context
Add any other context about the problem here.

[dongjuns]

How many images do you have?
for training and test.

[zhayanli]

50 thousands for training and 5 thousands for test

[dongjuns]

Thanks, and how many labels are in your training dataset?

[zhayanli]

90 thousands labels for 18 classes.

[RogerChern]

Could you please kindly share a part of your annotation and the prediction json files?

…

On Wed, Jun 17, 2020 at 10:15 AM zhayanli ***@***.***> wrote: 90 thousands labels for 18 classes. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#327 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABGODH3BPON6DG4EACAEQ5LRXARK7ANCNFSM4NFV442Q> .