how to read .dpt file?

[nita-xwj]

got a .dpt file,which is the ground-truth depth of 360 RGB scene, but when I use Hinterstousser to read dpt file, it goes wrong.Here is the code:

*```
INT_BYTES = 4
USHORT_BYTES = 2
def load_hinter_depth(path):
'''
Loads depth image from the '.dpt' format used by Hinterstoisser.
'''
f = open(path, 'rb')
h = struct.unpack('i', f.read(INT_BYTES))[0] # this output 1212500304
w = struct.unpack('i', f.read(INT_BYTES))[0] # this output 2048
depth_map = []
for i in range(h):
depth_map.append(struct.unpack(w'H', f.read(w * USHORT_BYTES)))
# return np.array(depth_map, np.uint16)
return np.array(depth_map, np.float32)

My 360 RGB PNG is 1024 height and 2048 width.When I change range(h) to range(1024), it can run, but the result array is obvious wrong.

[liuyuzhenn]

Hi, have you solved this?

[paulleem]

For anyone who's still wondering, here's my solution.

The .dpt file has 3 data of 4 bytes at the beginning.
The first one is an unknown data, following with width and height.

I was expecting to get a better result with this code, but turns out that the visualization depth image was quite a precise one.
Anyways, good luck to everyone.

import struct
import numpy as np

INT_BYTES = 4

def load_dpt_with_weird_header(path):
    with open(path, 'rb') as f:
        # Skip the first 4 bytes
        _ = f.read(INT_BYTES)

        # Read height and width
        w = struct.unpack('<i', f.read(INT_BYTES))[0]
        h = struct.unpack('<i', f.read(INT_BYTES))[0]

        # Read the pixel data (4 bytes each => 32-bit float)
        num_pixels = h * w
        pixel_bytes = f.read(num_pixels * 4)
        depth_vals = struct.unpack(f'<{num_pixels}f', pixel_bytes)

    # Reshape
    depth_map = np.array(depth_vals, dtype=np.float32).reshape((h, w))

    # Optional: normalize to [0, 1]
    min_val = depth_map.min()
    max_val = depth_map.max()
    depth_map = (depth_map - min_val) / (max_val - min_val)

    return depth_map