Daniel Bourke

thank you, going to test the changes and see what happens

Potential autolabelling pipeline: - raw images downloaded (e.g. filtered images from large dataset, such as, LAION-COCO) - several rounds of zero-shot classification are run to further filter images - "edible_food"...

See `openclip` for zero-shot classification: https://github.com/mlfoundations/open_clip Also see `clip-retrieval` for just embedding/searching a large existing dataset for images specific to a certain task: https://github.com/rom1504/clip-retrieval Can download a large number of...

Much better to compute image embeddings + class embeddings up front. Then reuse over time where necessary. This could be setup via: - image gets given UUID - image embedding...

See this resource for autolabelling object detection: https://github.com/facebookresearch/CutLER

See this notebook for an example conversion: https://github.com/mrdbourke/nutrify/blob/main/foodvision/notebooks/06_export_model.ipynb **Note:** Prediction/evaluation can only take place on macOS with CoreML, should setup some code to make a prediction over ~100 images with...

You're right I haven't seen a hook ability yet. Right now the ID's are manually sorted (e.g. I look through the database and see which ID best suits our food)....

Did this with a notebook and removed 695/25000 (or there abouts) images, saw a slight reduction in performance but this was expected due to less data leakage between train &...

Original notes (from #50) - * Found a library to help with image duplication thanks to hashing — https://github.com/idealo/imagededup Removing duplicates will help make the model more robust and prevent...

I should create some kind of simple app (Streamlit?) where a bunch of analytics about the labels/results of Nutrify are viewable. See Streamlit docs for connecting an app to Google...