I have recently initiated the training of Mask R-CNN for concrete particle detection and segmentation, specifically targeting coarse aggregates, voids, and cracks in X-ray CT grayscale scans. To build the model, I have relied on the architecture provided in this GitHub repository (https://github.com/Kamlesh364/Mask-RCNN-TF2.7.0-keras2.7.0).
Currently, I'm in the initial stages of model development, testing it on a limited dataset with only one epoch to assess instance prediction accuracy.enter image description here However, I've encountered an issue with the output images and predictions, which are illustrated in the images below. I also attempted to predict objects using the COCO class, but I consistently observe the same box pattern.
I would greatly appreciate it if someone could provide insights into the nature of this problem and its potential connections to other issues.
I have attempted to modify the configuration file to address this issue. enter image description hereenter image description hereBut I initially used the provided repository with a tested dataset. However, I encountered the problem I described earlier. Any guidance or insights on resolving this would be greatly appreciated.
