clvrai / Representation-Learning-by-Learning-to-Count

A Tensorflow implementation of Representation Learning by Learning to Count
MIT License
107 stars 16 forks source link

Resize features #3

Open ahmdtaha opened 4 years ago

ahmdtaha commented 4 years ago

Don't you need to resize AlexNet's feature before training the linear classifier? https://github.com/clvrai/Representation-Learning-by-Learning-to-Count/blob/2e86dfe22fabbbdaab18e858b835697b19f14b50/model_classifier.py#L109

In [1], check table #2 "the features of each layer are spatially resized until there are fewer than 9K dimensions left"

Also in [2], the author resize AlexNet features https://github.com/richzhang/colorization/blob/a1642d6ac6fc80fe08885edba34c166da09465f6/feature_learning_tests/linear/train_val.prototxt#L221

[1] Representation Learning-by-Learning-to-Count [2] Colorful Image Colorization