Semi automatic patching improves deep learning image segmentation accuracy and speed
Stride Independent Patching for Deep Learning
Computer Vision and Pattern Recognition
Summary
Segmenting images—dividing them into parts—is important for tasks like identifying objects. The paper introduces a method called SSP, which lets experts place patches on images to help deep learning models learn better. The authors compared SSP to other patching methods and found that SSP usually leads to better or comparable accuracy and faster training. This suggests expert-guided patch placement can improve model performance without longer training times.
What this means in practice
- •For medical imaging teams: Enhance organ or tumor segmentation model accuracy and reduce model training time by using expert-guided patch placement in medical images.
- •For automated quality control teams: Improve defect detection segmentation on product images by applying semi-automatic patching to focus on critical regions effectively.
Authors
Olivier Rukundo
Abstract
This paper presents semi-automatic stride-independent patching (SSP) as an alternative to automatic stride-dependent patching techniques. SSP uses user or expert input to position predefined patches over one or more objects of interest. To evaluate its effectiveness, three patch-based datasets were created using SSP, overlapping patching (Overlap), and non-overlapping patching (Noverlap). DeepLabV3+ models with ResNet50, ResNet18, and MobileNetV2 backbones were trained sepa-rately on each dataset. Quantitative evaluations were performed on the respective test splits and a common external test set. SSP generally achieved higher segmentation scores on the test splits and required the shortest model training time across all three backbones. On the external test set, SSP achieved the highest average precision and F1-score across backbones, whereas Noverlap achieved the highest average recall. These preliminary results demonstrate that the potentially greater spatial coverage of Noverlap and Overlap does not generally translate into better segmentation perfor-mance and that SSP offers a favorable balance between segmentation performance and model training time.