Hacker News new | ask | show | jobs
Using Tensorflow to Recognize Saimaa Ringed Seals from Webcam (blog.emblica.fi)
137 points by theikkila 3283 days ago
5 comments

Here is a tutorial on the easiest way to run Tensorflow: https://dcos.io/blog/2017/tutorial-deep-learning-with-tensor...
In case anyone is watching, the live stream is over :( [0].

Twitter account is located here [1].

[0] https://wwf.fi/en/norppalive/

[1] https://twitter.com/NorppaVahti

It was a good run while it lasted!

In case you guys have ideas for other similar ML/AI projects - we could take a look!

(shameless plug, I'm a partner at Emblica.fi)

Any idea where to find a database of birds? I've been kicking around with the idea of using transfer learning and then a database of bird images to fine tune a classifier to tell people what a bird they take a picture of is (EDIT, similar approach: [0]).

Would help with research. I know some people in finland that had participated in a crowdsourced census of birdlife in their area and thought it could be easier with an app that captures bird images and automatically classifies them (tells the user, I've often wondered what kind of bird something I've seen is) and provides the data to researchers.

I asked a friend of mine who works at a bird life charity if he knew of any data sets, but he didn't.

[0] https://jeffxtang.github.io/deep/learning,/tensorflow,/mobil...

The NABirds dataset: http://dl.allaboutbirds.org/nabirds.

But it's North American birds, not sure if you specifically want to cover birds that show up in Finland or if that was just your inspiration.

There are a couple apps that do what you're talking about, but they're both North American birds only as well.

http://www.birdsnap.com/ (I was involved in this one)

http://merlin.allaboutbirds.org/

I was looking into this for identifying other animals. iNaturalist has a community that adds observations and identifies species. They have an API as well as a dataset for research https://www.inaturalist.org/pages/developers
You could scrape something like this: https://www.reddit.com/r/whatsthisbird/
Is this related to that xkcd where he calls that task virtually impossible?
Does anyone know if there is any equivalent to a pre-trained network like Inception, but for text classification instead of image classification?
The closest is to use an LSTM with pretrained word embeddings like GloVe.
I'd love to see some sample code. I've tried something similar to detect the postman in my security camera stream, but failed over lots of small technical details.

Does anyone know a good tutorial on reinforcement learning that describes a practical application like this?

Yeah, the author of this article likely just followed these TensorFlow tutorials. You don't even have to touch code.

Follow this to retrain the Inception network: https://www.tensorflow.org/tutorials/image_retraining

Follow this to run an image through your model for classification: https://www.tensorflow.org/tutorials/image_recognition

I wanted to try out different types of final layers (SVM, Logistic Reg.) so the direct example didn't suit very well for that kind of testing out. For starters that good though.
Hello author of the article here!

I will probably publish in some time (when I'm not so busy doing other projects) some code samples too but until that I can give you tips.

You should start inspecting how the pre trained models (Inception v3/v4) works, what kind of layers they have and then decide what layers you want to use and what not. In case of Tensorflow, the tensorboard is very good tool for inspecting the model inner layers.

If you want to get started even more easier you should probably take a look into Tensorflow Slim models (https://github.com/tensorflow/models/tree/master/slim)

There is quite beginner friendly instructions for simple fine tuning of the models and it should take you pretty far.

The first lesson in the Fast.ai deep learning course covers exactly this (retraining a pre-trained network, in this case VGG, to distinguish between cats and dogs). There's a video and an accompanying ipython notebook with sample code, using Keras.

http://course.fast.ai/lessons/lesson1.html

OP said the following about AI SaaS providers:

You are using a “ready” product and so everyone gets the updates to model but your own modifications for the model are impossible with current providers.

I think this may not be true about Azure. In Build conference this year, they demoed cognitive service APIs where one can use transfer learning techniques, and train models and use them.

That sounds interesting, it's funny if you really can use their own models as base and do that. For the platform sake, Google offers also SaaS where you can train and evaluate your own models but then the base model is something you have to provide yourself

EDIT: I tried to google that up but couldn't find anything. Could you provide a link for that

Hmm ok, yeah apparently you can leverage their API's and teach new labels for your own data.

https://azure.microsoft.com/en-us/services/cognitive-service...

That is great and will definetly help with problems where your task isn't just to recognize cats and dogs the only downside is that you are giving your data away and it will also help your competitors.