search

axinc-ai / ailia-models-tflite

Quantized version of model library
21 stars 1 forks source link

Remove PIL from ESPCN #74

Closed kyakuno closed 7 months ago

kyakuno commented 7 months ago

ESPCNからPILを除去する。

kyakuno commented 7 months ago

最大誤差の計算

        img = Image.open(test_img_path)
        ycbcr = img.convert("YCbCr")
        y, cb, cr = ycbcr.split()

        img2 = cv2.imread(test_img_path)
        ycbcr = cv2.cvtColor(img2, cv2.COLOR_BGR2YUV)
        y2, cb2, cr2 = ycbcr[:,:,0], ycbcr[:,:,1], ycbcr[:,:,2]

        print(np.max(np.asarray(y).astype(np.float32) - y2),np.max(np.asarray(cb).astype(np.float32) - cb2),np.max(np.asarray(cr).astype(np.float32) - cr2))
0.0 3.0 -2.0