TropComplique / FaceBoxes-tensorflow

A fast face detector
MIT License
179 stars 66 forks source link

What is image input size? #18

Closed mrgloom closed 5 years ago

mrgloom commented 5 years ago

Is it right that model image input size is [None, 1024, 1024, 3] or it's just during training? i.e. at test time it can be [None,None,None,3]?

https://github.com/TropComplique/FaceBoxes-tensorflow/blob/1da7d8dd3dc816eeaf36082c9679a4be2fb70b11/train.py#L21 https://github.com/TropComplique/FaceBoxes-tensorflow/blob/9c4f6cb3a736620f64e3da91c0911586ccf354de/src/input_pipeline/pipeline.py#L28 https://github.com/TropComplique/FaceBoxes-tensorflow/blob/1da7d8dd3dc816eeaf36082c9679a4be2fb70b11/config.json#L20

mrgloom commented 5 years ago

Yes, it can have arbitary spartial size: https://github.com/TropComplique/FaceBoxes-tensorflow/blob/545ec4f4f3c55c3592ee189ed56a11a3fd017194/src/detector.py#L21

In paper they use VGA (640 x 480) image size at test time and 1024 x 1024 at train time.