New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
detection result #12
Comments
Hi, thanks for your intestest. Would you mind copy-pasting your training log (above the first itereation)? |
The log is too long, and I pase the log for the first epoch. 2021-10-20 19:13:26,408 - mmdet - INFO - Epoch [1][50/7330] lr: 9.890e-06, eta: 1 day, 16:17:48, time: 1.650, data_time: 0.880, memory: 3951, loss_cls: 1.2268, loss_bbox: 0.6944, loss: 1.9211 DONE (t=16.74s). |
I am sorry but I need the log information about loading the pretrained model, which is not included. |
Thanks a lot for your reply! |
Were the results improved in your training |
The AP has been improved to 38.6, the same as the reported value. |
Applying PVT detection framework, I tried a CycleMLP-B1 based detector with RetinaNet 1x.
I got AP=27.1, fairly inferior to the reported 38.6. Could you give some advices to reproduce the reported result?
The specific configure is as follows
base = [
'base/models/retinanet_r50_fpn.py',
'base/datasets/coco_detection.py',
'base/schedules/schedule_1x.py',
'base/default_runtime.py'
]
#optimizer
model = dict(
pretrained='./pretrained/CycleMLP_B1.pth',
backbone=dict(
type='CycleMLP_B1_feat',
style='pytorch'),
neck=dict(
type='FPN',
in_channels=[64, 128, 320, 512],
out_channels=256,
start_level=1,
add_extra_convs='on_input',
num_outs=5))
#optimizer
optimizer = dict(delete=True, type='AdamW', lr=0.0001, weight_decay=0.0001)
optimizer_config = dict(grad_clip=None)
find_unused_parameters = True
The text was updated successfully, but these errors were encountered: