Skip to content
This repository has been archived by the owner on Oct 31, 2023. It is now read-only.

About VGG16 pre-trained on ImageNet #36

Open
pengjw23 opened this issue Sep 26, 2022 · 3 comments
Open

About VGG16 pre-trained on ImageNet #36

pengjw23 opened this issue Sep 26, 2022 · 3 comments

Comments

@pengjw23
Copy link

we found that in the paper Chapter 4.2 : "ResNet101 [13] or VGG16 [36] pre-trained on ImageNet [7]". However, at adaptive_teacher/configs/faster_rcnn_VGG_cross_city.yaml, VGG16 did not used the pre-train ImageNet parameters like adaptive_teacher/configs/faster_rcnn_R101_cross_water.yaml.

We would like to know whether VGG16 are pretrained on ImageNet or not. thank you very much

@yujheli
Copy link
Contributor

yujheli commented Sep 30, 2022

Good question. Our model using VGG16 is trained from scratch.

@pengjw23
Copy link
Author

pengjw23 commented Oct 8, 2022

Thank you for your reply, and I want to know something about the dataset Foggy Cityscapes. I want to know which level weather you use when training, or all three types of weather are used for training. As a newcomer to this field, I am confused about this problem. I noticed that some papers only use foggy weather with level=0.02 for evaluation, for example, in Table 1 in PT (ICML-22, [1]).

1665199676458

[1]Chen M, Chen W, Yang S, et al. Learning Domain Adaptive Object Detection with Probabilistic Teacher[C]//International Conference on Machine Learning. PMLR, 2022: 3040-3055.

@Manjuphoenix
Copy link

@yujheli

Good question. Our model using VGG16 is trained from scratch.

So the parameter DATASETS.TRAIN inside Base-RCNN-C4.yaml file has no effect in the training process or does it use faster_RCNN model use the model pretrained on coco_2017_train dataset that is registered in the detectron?

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants