Results from Cell-TRACTR and model overview for object detection.
<p>(A) Representative output from Cell-TRACTR for a single chamber from a time-lapse microscopy movie of <i>E. coli</i> bacteria growing in the mother machine microfluidic device. The output is displayed as a kymograph where the same chamber is shown over time. Each color represent...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Published: |
2025
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | <p>(A) Representative output from Cell-TRACTR for a single chamber from a time-lapse microscopy movie of <i>E. coli</i> bacteria growing in the mother machine microfluidic device. The output is displayed as a kymograph where the same chamber is shown over time. Each color represents a unique cell being tracked and black arrows signify cell divisions. This movie is from the test set. (B) Overview of Cell-TRACTR tracking mammalian nuclei across multiple frames. Each color represents a unique cell being tracked. These frames are from the DeepCell dataset using data from the test set. (C) Cell-TRACTR is a DETR-based model that first uses a CNN backbone to extract image features, then uses a transformer encoder-decoder framework that leverages attention mechanisms. Encoded multi-scale features are used for query selection to initialize the object queries. Object queries attend to the encoded multi-scale features to generate the output embeddings needed to predict class labels, bounding boxes, and segmentation masks. If the model predicts an object query to be a cell, then the output embedding is shown as a colored box in the schematic. Object queries predicted to be of class “no object” are colored black and are discarded. The squares represent content embeddings and the circles represent positional embeddings.</p> |
|---|