richardaecn / cvpr18-inaturalist-transfer Goto Github PK

NABIRDS datasets are showing very bad result in terms of accuracy (31.75%). Any suggestion, why?

I have been stay tuned for months.

Is there a code that can reproduce the results in the author's paper?

Hi,
Thanks for your work. Do you have the tf.SavedModel version of the pretrained models you show ?
If not, is there a way to transform a ckpt file in a tf.SavedModel easily ? I would like to use your model in keras at the very end if possible.

Amaury

https://drive.google.com/file/d/1vOHKuqt7XgROo9t5cblJvGf0kGpiFaF1/edit cannot open ?is it right?

Hi, I Have a couple of questions:

1. In LinearClassifierDemo.ipynb, you gave the query image and it extracts similar images from training data. What is the criteria that extract those five images? Like is there any feature matching criteria?

2. The features we got by using pretrained models, these pretrained models are available publicly or you trained these by yourself and which pretrained model train on which dataset?

How to get the subset A/B in Table 6? Can you share the subset A and subset B ?

Hi, is it possible to provide the JFT-pretrained ResNet-101 checkpoint so we could calculate the domain similarity for other custom datasets?