TRI-ML
Running inference with visualization
Hi there, I'm trying to test out the pre-trained model on a custom dataset, but I am unable to find the script that does that (Seems like visualize_dataloader is visualizing the GT). I've been trying to understand the hydra experiments but I am not sure if there is an option to do just prediction without ground truths.
Any help on this would be highly appreciated, thanks!
ps. this is assuming I have the camera parameters
Hi @amirhmk, out of the box, I don't think the repo comes with that ability. However, you can provide empty dummy ground truth with your custom data and run inference. This would potentially enable you to do what you wanted without fiddling around much with the detectron framework. Also, I think the repo saves prediction's visualization as images in compressed .npz format (compresses the image array). You can potentially reconstruct the images from the .npz array and visualize them with opencv or PIL
