4 channel Data in 3 channel pretrained model https://blog.naver.com/hahmwj/222854175154
I tuned pretrained model' input size in tensorflow(res50, vgg16) but my data's shape is (28,28,4).
Solution1 Fine tuning Transfer Model 3 to 4 I tryed to fin tune model's input size 3 channels to 4channels and (28, 28, 3) to (32, 32, 3). But it's acc is to low.
Solution2 Data Squeeze 4 channel to 3channel v2.ipynb So I transpose mu dataset to 4 3 channels Datasets finally it got 4 kinds of model that shape (28, 28, 3) and use augmentation method by ImageGenretor by tensorflow. Also, I used ZeroPadding method to transpose (28, 28, 3) to (32, 32, 3).
For evlaution For test Last Model.ipynb First, I transpose my test dataset (28, 28, 4) to (32, 32, 3) either upper one, Second, I make 4 * len(model count) condition. and evaluate each other then choose best acc in the array.