I am using TLT to train a classifier(resnet18) for OCR. I have 3 queries.
- How should the labels.txt in deepstream-test_app-2 look for a given classmap.json?
Here is my classmap.json
{"1": 1, "0": 0, "3": 3, "2": 2, "5": 5, "4": 4, "7": 7, "6": 6, "9": 9, "8": 8, "A": 10, "C": 12, "B": 11, "E": 14, "D": 13, "G": 16, "F": 15, "H": 17, "K": 19, "J": 18, "M": 21, "L": 20, "O": 23, "N": 22, "Q": 25, "P": 24, "S": 27, "R": 26, "U": 29, "T": 28, "W": 31, "V": 30, "Y": 33, "X": 32, "Z": 34}
And this is my labels.txt
0;1;2;3;4;5;6;7;8;9;A;B;C;D;E;F;G;H;J;K;L;M;N;O;P;Q;R;S;T;U;V;W;X;Y;Z
.
But my results are quite random even though my validation accuracy was 99 during training. So the .engine file shouldnt be the problem. How should my labels.txt be? Can it have numbers? Is only lowercase allowed?
-
Does the input size have to be a square? I get errors during evaluation if I try with different height and width i.e, has to be 224,224 and not 112,224.
-
Is Data Augmentation possible for classification on TLT? If yes where to specify the parameters on the spec.cfg file?