FLIM Networks with Bag of Feature Points

TL;DR

This study revisits FLIM SOD and introduces FLIM-Bag of Feature Points (FLIM-BoFP), a considerably faster filter estimation method compared to FLIM-Cluster and other state-of-the-art baselines for parasite detection in optical microscopy images.

Abstract

Convolutional networks require extensive image annotation, which can be costly and time-consuming. Feature Learning from Image Markers (FLIM) tackles this challenge by estimating encoder filters (i.e., kernel weights) from user-drawn markers on discriminative regions of a few representative images without traditional optimization. Such an encoder combined with an adaptive decoder comprises a FLIM network fully trained without backpropagation. Prior research has demonstrated their effectiveness in Salient Object Detection (SOD), being significantly lighter than existing lightweight models. This study revisits FLIM SOD and introduces FLIM-Bag of Feature Points (FLIM-BoFP), a considerably faster filter estimation method. The previous approach, FLIM-Cluster, derives filters through patch clustering at each encoder's block, leading to computational overhead and reduced control over filter locations. FLIM-BoFP streamlines this process by performing a single clustering at the input block, creating a bag of feature points, and defining filters directly from mapped feature points across all blocks. The paper evaluates the benefits in efficiency, effectiveness, and generalization of FLIM-BoFP compared to FLIM-Cluster and other state-of-the-art baselines for parasite detection in optical microscopy images.

Figures (5)

• Figure 1: Training pipeline of a FLIM SOD Network from user-drawn markers on three representative images. In this example, the markers are red disks on the object (Schistosoma Mansoni eggs) and white disks on the background (impurities and empty region).
• Figure 2: Progressive saliency maps by decoding consecutive blocks. Figures \ref{['original']} and \ref{['truelabel']} show the original and groundtruth images, respectively. Saliency maps obtained by decoding the output of blocks 1-4 for FLIM-BoFP in \ref{['bofp1']}-\ref{['bofp4']} and FLIM-Cluster in \ref{['clust1']}-\ref{['clust4']}.
• Figure 3: Training of a FLIM encoder with BoFP begins with a single clustering-based feature point estimation to insert discriminative image locations in the BoFP. Filters for each block are then directly derived from mapped points in the feature map at the block’s input, ensuring precise filter placement and efficient processing.
• Figure 4: Average F-measure curves across splits on test sets for each model. Figure \ref{['fig:before_after_post']} compares FLIM models with and without DT post-processing for S. Mansoni. Figure \ref{['fig:compare_dt']} contrasts FLIM$_{DT}$ models against deep-learning models on S. Mansoni, with the FLIM models at the top.
• Figure 5: Qualitative detection results on test images of the S. Mansoni, Entamoeba and Ancylostoma datasets. GT is the ground-truth segmentation mask.