search

isl-org / DPT

Dense Prediction Transformers
MIT License
1.97k stars 256 forks source link

Metric Depth from Kitti model #81

Open talasalim opened 1 year ago

talasalim commented 1 year ago

Hi there, I have ran the monocular depth estimation model using the mono-depth fine-tuned on Kitti on one one of my images: python run_monodepth.py --model_type=dpt_hybrid_kitti

I want to extract the metric depth at every specified pixel so I wrote this code:

img = cv2.imread("results25_kitti2.pfm", cv2.IMREAD_UNCHANGED)

height, width = img.shape[:2]

x = int(input("Enter x coordinate: "))
y = int(input("Enter y coordinate: "))

if x < 0 or x >= width or y < 0 or y >= height:
    print("Invalid coordinates")
else:
    pixel_value = img[y, x]
    print("Pixel value:", pixel_value)

However I'm getting values in the range of 2000-4000 and I'm not sure whether they are the metric depth in cms?

hoangthang1607 commented 1 year ago

Where did you take the results25_kitti2.pfm file? Normally, depth map from KITTI is written in uint16, and you should divise by 256 to get depth values in meters.