the pre-trained model
woqiaow opened this issue · comments
Hello, thank you for sharing the code. I have some problems in implementing the code.
In voc.yaml, the pre-trained caffemodel is data/models/coco/deeplabv1_resnet101/caffemodel/deeplabv1_resnet101-coco.pth, I want to ask whether this pre training model is used for training voc dataset?
In addition, if I want to replace it with the ImageNet pre-trained model, can I directly change the path of the deeplabv1_resnet101-coco.pth to the path of the ImageNet pre_trained model in voc.yaml?
I hope you can answer, thank you!
Q: The pre-trained model is used for training VOC?
Yes, the official DeepLab v2 is also initialized on COCO and fine-tuned on Pascal VOC.
deeplabv1_resnet101-coco.pth
is converted from the officially provided init.caffemodel
.
Q: Can I directly change the path to replace the init model?
Yes, it's okay if your ImageNet weights are compatible with this repository re-implemented from the Caffe ResNet. For instance, torchvision ResNet is not compatible (#61, #67, #88).
You have an option of the on-the-fly modification (#33).
Thank you for your quick reply! Very clear.