hardnet85_base.pth
Hi, How is hardnet85_base.pth derived? It seems like the backbone is different from the one used for classification. Thanks.
Hi, the base model is from SSD-HarDNet85, and since the model for SSD is slightly different from the classification one it was trained with ImageNet and early termination (stop on ~120 epochs / 150 epochs)
Hello, @PingoLH
Then, how do I use that hardnet85_base.pth??
When I use this command, it doesn't work.
python main.py ctdet --exp_id coco --dataset coco --arch hardnet_85 --gpus 0 --num_epochs 100 --lr_step 60,80 --load_model hardnet85_base.pth
I've changed some model loading codes and debug it, but state_dict of pth file is different to the hardnet model in your repository. (The length of dict is also different!)
I want to know how to use those hardnetXX_base.pth files in your repository 🙏