DCTNet icon indicating copy to clipboard operation
DCTNet copied to clipboard

jpeg2dct or cv2.dct

Open YuqiangY opened this issue 4 years ago • 5 comments

This is a wonderful research idea and thought process. But I have a question now: Are there many differences between jpeg2dct and cv2.dct with blockwise? I have used them to do classification on Cifar10, but they have the same results almostly.

YuqiangY avatar Aug 28 '20 06:08 YuqiangY

您好,请问cv2.dct 具体是怎么操作的呀?

DayBreak-u avatar Dec 30 '20 01:12 DayBreak-u

您好,请问cv2.dct 具体是怎么操作的呀?

貌似是利用到了fft,反正两者计算结果是有差别的,但是最终的效果差不多

YuqiangY avatar Jan 04 '21 16:01 YuqiangY

您好,jpeg2dct后得到的dct_y、dct_cr、dct_cb ,他是(M,N,64)大小的,这个64维度是不是按照低频到高频排列的,有没有ZigZag呢?

luowei0701 avatar Jan 09 '21 08:01 luowei0701

@LBJ23233 请问你利用OpenCV实现的,有对DCT输入进行Normalize进行归一化吗?

J-JunChen avatar Oct 15 '21 06:10 J-JunChen

Why does the model trained using data returned from jpeg2dct have such poor accuracy? It only achieves 66% accuracy on CIFAR-100

B-JackMao avatar Jul 04 '24 09:07 B-JackMao