DDRNet-23-slim, cannot reproduce results.

[rawalkhirodkar]

Hello, Thank you for the great work!

I am trying to reproduce the results of ddrnet-23_slim on Cityscapes val (77.83 mIoU) but could not do it.

• I tried training for more than 484 epochs - 600 epochs and 1500 epochs with 2, 5 and 8 GPUs.
• Evaluating with both heads model index 0 and model index 1
• The best performance is around 74.2 mIoU.
• I also noticed, the results do not match when you use different number of gpus but train for the same number of epochs.

Here are the training logs, 600epochs_5gpus_ddrnet23_slim_2021-06-02-20-21_train.txt 600epochs_8gpu_ddrnet23_slim_2021-06-03-09-32_train.txt 1500epochs_2gus_ddrnet23_slim_2021-06-04-18-36_train.txt

I would really appreciate your help. Thank you.

[ShihuaHuang95]

@rawalkhirodkar I got 75.6, but the performance is still worse than the stated.

[ShihuaHuang95]

Besides, the reproduced result of DDRNet-23 is low.

[chenjun2hao]

@rawalkhirodkar @ShihuaHuang95 ，i just train with 2*3080 gpus with the config files. i use the official imagenet pretrained model. Nothing else has been changed.

[rawalkhirodkar]

I think this has to do with the learning rate updates. num_iters_per_epoch is a function of number of gpus which in turn controls the learning rate decay.

Training for longer (1500 epochs) with 8 GPUs reproduces the OWN performance with FLIP mentioned in the README.

Thanks.

[YAwei666]

I think this has to do with the learning rate updates. num_iters_per_epoch is a function of number of gpus which in turn controls the learning rate decay.

Training for longer (1500 epochs) with 8 GPUs reproduces the OWN performance with FLIP mentioned in the README.

Thanks.

Are you use the same config, when train 1500 epoch with 8 GPUs reproduces the OWN performance? or you have changed something? I would really appreciate your help. Thank you.

[chenjun2hao]

这是来自QQ邮箱的假期自动回复邮件。   您好，我最近正在休假中，无法亲自回复您的邮件。我将在假期结束后，尽快给您回复。