models

Google Photos is the wrong product to ask this question, IMHO. One of its key highlight is sharing albums with others and preserving them for the future, so why waste a consumer device's energy on unnecessary tasks if the image is going to end in a Google datacenter anyway? Google Lens, which isn't about sharing, does run many functions on device: for example, it does ondevice offline translation: https://9to5google.com/2021/01/25/google-lens-translate-offl... The general direction at Google (https://ai.googleblog.com/2021/11/improved-on-device-ml-on-p...) is to move as much as possible on device for privacy and latency reasons. There are published and opensource (Apache license, according to the github LICENSE file) models for Tensorflow Lite which runs on Pixel Edge TPUs (and most embedded devices): https://tfhub.dev/s?q=edgetpu https:/

Fresh and (I think) clean implementations of Faster R-CNN in PyTorch and TensorFlow 2/Keras. I wanted to learn about object detectors and decided to understand and implement a foundational model in the field, Faster R-CNN (elements of which are still used in modern models to this day), using the paper alone. That proved to be more difficult than expected and I had to relent and take a peak at existing implementations to fill in some important gaps. I've documented my struggles and learnings in the README for others to benefit from. I also wanted to solidify my understanding of TensorFlow/Keras and learn PyTorch, so I made sure to implement the model in both. Faster R-CNN is fairly challenging in that it doesn't quite map to Keras tutorial examples. For example, losses are not computed simply as a function of model output and input but rat

I just tried "UnsafeVarint" in DDG and while #1 and #2 were useless COVID hits as you said, DDG #4 was the same as Google's #1 hit which was this GitHub issue: https://github.com/tensorflow/models/issues/7420 Google's #2 hit for me was blogspam at "bankoftrans [dot] com" that copied the source code for UnsafeVarint in order to game Google search results. (A growing problem others in this discussion have been complaining about.) So Google isn't stellar for your search, either, though perhaps it was working better for you because you were logged in with your Google Account. Like I said, I do still use Google via "!g" and "!gi" when DDG's results are garbage, but often in these cases Google's results are also junk.

For the purpose of this question, you might search for pre-trained models that predict what you want to predict, and use them. https://modelzoo.co/ https://github.com/tensorflow/models https://cv.gluon.ai/model_zoo/index.html If there are no pre-trained models, you could take a look at fast.ai courses: https://course18.fast.ai/ml.html (for Machine Learning (ML)) https://www.fast.ai (for others "DL" (Deep Learning)) This courses is targeted towards programmers, and they dive right into the code to use ML (train models and use them) with Jupyter notebooks. They also have a book whose notebooks are here: https://github.com/fastai/fastbook If you have some maths background (high-school, or first year university), check out Stanford's CS229 with Andrew Ng: https://www.youtube.com/playlist?list=PLoROMvodv4rMiGQp3WXSh... I think it's better than

Yes, most of the large models that are not feasible to be trained by individuals are downloadable. Frameworks make it easy to import and export snapshots of model weights. See https://github.com/tensorflow/models/tree/master/research/sl...

How does this compare to https://github.com/tensorflow/models/blob/master/research/ob... ? (Tensorflow is not needed for some of those functions)

Meh - it didn't work that well when I tried it. [1] Params I used were {architecture: 'ResNet50', outputStride: 32, quantBytes: 2}, with segmentMultiPerson function. The same images DeepLab handles perfectly. [2] If anyone sees a problem with the params/method I used let me know. [1] https://superanimo.com/bodypix/ [2] Deeplab: https://github.com/tensorflow/models/tree/master/research/de...

I work a lot with object detection at my current role. Would be cool to add Detectron 2, seems that this library that has some of the state of the art(SOTA) models in object detection. https://github.com/facebookresearch/detectron2 Also the TF Object Detection API has a suite of pretrained SOTA object detection models https://github.com/tensorflow/models/tree/master/research/ob... What makes object detection really valuable is being able to finetune a model to detect objects that are relevant to your problem. However collecting, labeling, takes a lot of time since the best way is to manually label. But 30-50 hrs of manual labeling can get you some great results when you are leveraging pretrained models. Building a tool to manage data labeling and training for finetuning object detection models would be very valuable.

What do you mean? Basically all Neural-network models (from Google and others) are freely available. And to top that you can also get pre-trained weights for the models for free. https://www.tensorflow.org/hub/ https://github.com/tensorflow/models

For detecting objects in video, take a look at: https://github.com/tensorflow/models/tree/master/research/ls... It uses SSD object detector (https://arxiv.org/abs/1512.02325 , https://lambdalabs.com/blog/how-to-implement-ssd-object-dete...) on each frame, then runs LSTM on top of it to collect information across video frames. The trained model can be run on mobile devices via Tflite.

There are probably a bunch on ProductHunt. For example https://remove.bg got a lot of upvotes - it probably uses this on the backend: https://github.com/tensorflow/models/tree/master/research/de...

If you're interested, this project is open source! https://github.com/tensorflow/models/tree/master/research/as... https://ai.googleblog.com/2018/03/open-sourcing-hunt-for-exo...