search

ydhongHIT / DDRNet

The official implementation of "Deep Dual-resolution Networks for Real-time and Accurate Semantic Segmentation of Road Scenes"
MIT License
402 stars 54 forks source link

Saved model without pretraining #26

Closed hojunson closed 1 year ago

hojunson commented 2 years ago

Hello Yuanduo,

I think your achievement is really awesome. I am looking into your code and your model to develop more on it.

I was wondering whether the "best.pth" model is the trained model without Imagenet pre-training.

I remember I got the file from the google drive and it shows about 75.xx in validation set of Cityscapes.

Could you confirm it?

ydhongHIT commented 2 years ago

Hello Yuanduo,

I think your achievement is really awesome. I am looking into your code and your model to develop more on it.

I was wondering whether the "best.pth" model is the trained model without Imagenet pre-training.

I remember I got the file from the google drive and it shows about 75.xx in validation set of Cityscapes.

Could you confirm it?

All the models utilize the imagenet pre-training. 75.xx is abnormal. There are some third-party repos or papers reproducing my method and it is easy to achieve above 77% mIoU for ddrnet_23_slim. The model I provided should be 77.8.

ydhongHIT commented 2 years ago

Hello Yuanduo,

I think your achievement is really awesome. I am looking into your code and your model to develop more on it.

I was wondering whether the "best.pth" model is the trained model without Imagenet pre-training.

I remember I got the file from the google drive and it shows about 75.xx in validation set of Cityscapes.

Could you confirm it?

If you have some problems, you can refer to this repo which provides the eval script.