Closed agentmorris closed 1 year ago
All of the .json files for camera trap datasets on LILA are in COCO Camera Traps format. Bounding boxes are [x,y,w,h], where:
Let us know if you find anything that isn't consistent with that.
Hope that helps!
(Comment originally posted by agentmorris)
Need clarifications on the conventions used for bounding box parameters in JSON [x, y, w, h].
I am currently using the dataset that was downloaded from the below link. https://lilablobssc.blob.core.windows.net/caltechcameratraps/eccv_18_all_images_sm.tar.gz
Assumed x, y as centers of the bounding box. However, when I try to normalize, I get values >1 or sometimes <0 also.
Please find the attached spreadsheet for reference. train_normalized.xlsx
Here's how I went about calculating. Taking the example for image_id = '586fde22-23d2-11e8-a6a3-ec086b02610b'
Bounding box values as given in the json. x_center = 1254.455625 y_center = 591.602857142857 w = 175.361428571429 h = 156.157053571429
Would want to normalize these values and bring them within [0-1].
-# Image width and height values -# Found these values using imagesize -# imW, imH = imagesize.get(src_path) imW = 1024 imH = 747
-# top-left coordinate of the box xmin = x_center - w/2 = 1166.77491071429 xmin_normal = xmin/imW = 1.13942862374442 # > 1
I am guessing (x,y) to be one of the corners of the bounding box. Can you please clarify.
Issue cloned from Microsoft/CameraTraps, original issue posted by ra9hur on Oct 13, 2022.