tensorflow / tfhub.dev

TensorFlow Hub has moved to Kaggle Models

Starting November 15th 2023, links to tfhub.dev redirect to their counterparts on Kaggle Models. tensorflow_hub will continue to support downloading models that were initially uploaded to tfhub.dev via e.g. hub.load("https://tfhub.dev/<publisher>/<model>/<version>"). Although no migration or code rewrites are explicitly required, we recommend replacing tfhub.dev links with their Kaggle Models counterparts to improve code health and debuggability. See FAQs here.

As of March 18, 2024, unmigrated model assets (see list below) were deleted and retrieval is no longer possible. These unmigrated model assets include:

• inaturalist/vision/embedder/inaturalist_V2
• nvidia/unet/industrial/class_1
• nvidia/unet/industrial/class_2
• nvidia/unet/industrial/class_3
• nvidia/unet/industrial/class_4
• nvidia/unet/industrial/class_5
• nvidia/unet/industrial/class_6
• nvidia/unet/industrial/class_7
• nvidia/unet/industrial/class_8
• nvidia/unet/industrial/class_9
• nvidia/unet/industrial/class_10
• silero/silero-stt/de
• silero/silero-stt/en
• silero/silero-stt/es
• svampeatlas/vision/classifier/fungi_mobile_V1
• svampeatlas/vision/embedder/fungi_V2

Thank you for using tfhub.dev over the years and see you at Kaggle Models!

tfhub.dev's Issues

Regarding replacing a Hub model

I am writing to know what's the best way to replace a model on TensorFlow Hub? Today, I noticed the int8 EAST model is a bit buggy and I worked on a fix today. I wish to replace the existing one now. What should be the course of action here?

Shortening the ViT model space

@MorganR @kempy here is the search criteria I used to shorten the ViT model space -

1. Top-1 accuracy on the ImageNet-1k val set should be at least 75%.
2. Fileterd out the models having an input size of 224 x 224.
3. Then I reverse sorted all the models from 1) and took the top-10.

The top-1 accuracies we got from these models are bounded within [77.002%, 79.836%]. The models (S/16) we get from the above have ~22 Million parameters. This is comparable to a standard Resnet50.

Note that these models were first pre-trained on the ImageNet-21k dataset and then were fine-tuned on ImageNet-1k. When I applied the similar conditions on the models pre-trained on ImageNet-1k directly, the highest top-1 accuracy I got on the ImageNet-1k val set is 76.696%.

Let me know if these 10 models look good to you and I will then proceed with the rest of the steps.

My model exploration notebook is available here.

Publisher page not found

Hello,

I just noticed that when I go to the publisher pages: https://tfhub.dev/s?subtype=publisher
the link to the publisher page of the cosmo-group-ethz: https://tfhub.dev/cosmo-group-ethz
throws a 404. Is there something wrong with the cosmo-group-ethz.md that we submitted recently or is this related to something else?

Dataset used for pretraining and finetuning GTR model

I'm trying to reproduce the results presented in Large Dual Encoders Are Generalizable Retrievers paper and perform some experiments.
So I'm in need of the dataset used for pretraining and finetuning GTR models.

Adding a plant disease model

I am planning on adding TensorFlow SavedModel and TFJS models for identifying plant diseases using the plant_village dataset on top of MobileNet. My final set of experiments for the models can be found in this repo. I did not find any such model or related model on TF Hub.

I request you to kindly review this, should I create a PR adding this model?

the labelmap.txt file is missing in your yolo model on kaggle

Hello, i tried to employ your yolo v5 model, but there are a mssing file, where the labels are named.

What can i do?

Somewhere i read, that the label files are included into the tensorflow lite models. (.tflite)

My list from a other model is in the same file like the .tflite model an included something like this:

cannot push to this repo.

I made an edit on a branch named fix/type-dataset-name derived from master to issue a PR and git push origin fix/type-dataset-name. Then I got the following error:

$ git push origin fix/type-dataset-name 
ERROR: Permission to tensorflow/tfhub.dev.git denied to DameNianch.
fatal: Could not read from remote repository.

Please make sure you have the correct access rights
and the repository exists.

It's strange that I can push to my personal private github repository. Does this repository have permission for anyone to create new branches?Only Contributors and Collaborators?

Packaging a SavedModel

Hi there.

How a SavedModel should be packaged?