Hacker News new | past | comments | ask | show | jobs | submit login
Simple Python Package to Extract Deep Learning Features (chsasank.github.io)
91 points by chsasank 43 days ago | hide | past | web | favorite | 11 comments



This blogpost makes it sound as if using a deep learning library for a task like image classification is difficult, while it is just a matter of copy+paste, annotate, and run.


Dunno, maybe you aren't the target for the article. As someone who has never played with deep learning, I thought it was presented just fine.


it is a lot harder than what this library does.


Please provide either an example or a source for this assertion.



The OP library specifically grabs the ImageNet model and uses transfer learning. That's just a general keras convolutional networks resource.


The entire library is only 100 lines of code, so you could copy and paste its own source.


I wonder if there is a BERT equivalent for NLP. Is BERT part of Spacy now?



The Tf-Hub stuff is pretty close I guess https://tfhub.dev/google/bert_uncased_L-12_H-768_A-12/1


huggingface offers a whole family of pretrained transformers.

https://github.com/huggingface/pytorch-transformers




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: