python - 为什么我的numpy文件比使用同一数组生成的PNG大？

Question

我需要找到一种可以容纳我的数组而不占用太多内存的文件类型。我以为npy文件会小于PNG，但我想我错了。

            depth_img = np.uint16(depth * 256)

            cv2.imwrite(name_dest_im, depth_img)
            np.save(name_dest_npy, depth_img)

是因为我在导出数组时做错了什么，还是PNG是存储uint16的最有效方法？
编辑:这是 https://github.com/nianticlabs/monodepth2/blob/master/test_simple.py的修改的版本

input_image = pil.open(image_path).convert('RGB')
            original_width, original_height = input_image.size
            input_image = input_image.resize((feed_width, feed_height), pil.LANCZOS)
            input_image = transforms.ToTensor()(input_image).unsqueeze(0)

            # PREDICTION
            input_image = input_image.to(device)
            features = encoder(input_image)
            outputs = depth_decoder(features)

            disp = outputs[("disp", 0)]
            disp_resized = torch.nn.functional.interpolate(
                disp, (original_height, original_width), mode="bilinear", align_corners=False)

            # Saving numpy file
            output_name = os.path.splitext(os.path.basename(image_path))[0]
            name_dest_npy = os.path.join(output_directory, "{}_disp.npy".format(output_name))
            scaled_disp, _ = disp_to_depth(disp, 0.1, 100)


            name_dest_im = os.path.join(output_directory, "{}_disp.png".format(output_name))
            # Saving colormapped depth image
            map_resized_np = scaled_disp.squeeze().cpu().numpy()
            map_resized_np = cv2.resize(map_resized_np, (original_width, original_height))
            depth = 5.4 / map_resized_np
            depth = np.clip(depth, 0, 80)
            depth_img = np.uint16(depth * 256)

            cv2.imwrite(name_dest_im, depth_img)
            np.savez_compressed(name_dest_npy, depth_img)

Answer 1

PNG被压缩，但是np.save不使用压缩。您可以使用np.savez_compressed保存压缩数组。如果您的数据允许，您也可以另存为uint8，如@vlovero在评论中建议的那样。

import cv2
import numpy as np

img = np.random.randint(0, 256, size=(128, 128, 3)).astype(np.uint16)

cv2.imwrite("random.png", img)
np.save("random.npy", img)
np.savez_compressed("random.npz", img=img)

f = np.load("random.npz")
img_loaded = f["img"]
# The loaded array is equal to the original.
np.testing.assert_array_equal(img, img_loaded)

文件大小如下:

$ du -h random*
100K    random.npy
64K random.npz
64K random.png