indoor-outdoor

Indoor/Outdoor Classifier

This simple classifier can identify if an image is take indoors or outdoors. This is a model developed to test a hypothesis that such a classification can be made using 3 things:

• very limited data set
• transfer learning from a small model
• very quick inference to keep deployment costs low

Performance

In the training that I completed for this model, I have the following:

• Training set = 65 images (around 32 images for each label)
• Validation set = 16 images (8 images for each label)
• Testing set = 19 images (9 images for each label)

This is definitely sub-optimal. However, I want to use this to test my hypothesis. So this is good enough as a first iteration.

Pre-Trained Model

For this use-case, I am using the pre-trained model MobileNetV2. In my use case, I have just 50 images each for the 2 labels that I will be training. Due to the small data set, a smaller model like MobileNet would be better suited.

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference