RameezI / dishscan

This repository explores AI/ML approaches towards automated tracking of dietary intake: estimating the ingredients & macro-nutrients in a food plate using digital images.
MIT License
2 stars 0 forks source link

pre-trained weights #1

Closed nadiaparsell closed 1 month ago

nadiaparsell commented 2 months ago

Hello,

I was wondering if you have pre-trained weights we can test the model with? Thanks!

nadiaparsell commented 2 months ago

Also, what is the directory structure for the images? the dataset is quite big so I want to try running it with the dummy dataset but the script tries to automatically download the entire nutrition5k dataset even when I set the dataset path to dishscan/datasets/nutrition5k/dummy_data/nutrition5k_dataset

RameezI commented 1 month ago

Hi @nadiaparsell,

You can use the pre-trained MobileNetV2 model for testing. To do this, follow these steps:

  1. Set up the environment by completing steps 1-3 in the README.md.

  2. Download the .zip package containing the pretrained model. MobileNetV2.zip

  3. Unzip the model directory and run the classification script as shown below:

unzip MobileNetV2.zip
python scripts/classify_image_tf.py -m MobileNetV2 -i test_images/dish_1551374783_A.png

For more information on the dataset, please see: https://github.com/google-research-datasets/Nutrition5k

nadiaparsell commented 1 month ago

Thank you!