Describe the bug
Prediction on horizontally biased images (long thin along the x-axis) don't predict properly using deepblink.inference.predict.
To Reproduce
Steps to reproduce the behavior:
Using the image below:
Import both image and model and run pink.inference.predict(image, model) and it will only return 0.0 coordinates.
How to solve
The image used has a padded shape of (64, 512). The current implementation of deepblink.inference.predict uses deepblink.data.get_coordinate_list(pred, image_pad.shape[0]). This [0] will lead to issues as the coordinate conversion will discard all coordinates beyond (64, 64). Using max(image_pad.shape) instead should do the trick.
Describe the bug Prediction on horizontally biased images (long thin along the x-axis) don't predict properly using
deepblink.inference.predict.To Reproduce Steps to reproduce the behavior:
pink.inference.predict(image, model)and it will only return 0.0 coordinates.How to solve The image used has a padded shape of (64, 512). The current implementation of
deepblink.inference.predictusesdeepblink.data.get_coordinate_list(pred, image_pad.shape[0]). This [0] will lead to issues as the coordinate conversion will discard all coordinates beyond (64, 64). Usingmax(image_pad.shape)instead should do the trick.