Need for resizing before evaluating

yeephycho / tensorflow-face-detection

A mobilenet SSD based face detector, powered by tensorflow object detection api, trained by WIDERFACE dataset.

Apache License 2.0

768 stars 264 forks source link

Need for resizing before evaluating #45

Closed bostafh closed 5 years ago

bostafh commented 5 years ago

Hello, Is the model performing some resizing in the tensorflow back-end ? I get significantly better result when i crop the outside of an image to make it square. So could it be that the model is resizing images to make them square and by doing so squishing faces ? Thanks

yeephycho commented 5 years ago

Keep the feature of a face will help to get a better detection result.

bostafh commented 5 years ago

What does that mean sorry ? You're not addressing my question, are you ? You can try by yourself but results are way better on square images than in original aspect ratio.