CholecT40 (Cholecystectomy Action Triplet)
Introduced by Nwoye et al. in Recognition of Instrument-Tissue Interactions in Endoscopic Videos via Action Triplets
CholecT40 is the first endoscopic dataset introduced to enable research on fine-grained action recognition in laparoscopic surgery.
It consists of 40 videos of laparoscopic cholecystectomy surgery annotated with triplet information in the form of <instrument, verb, target>. The annotations spans over 128 triplet classes that are composed from 6 classes of surgical instruments, 8 classes of action verbs, and 19 classes of surgical targets.
The dataset is used as benchmark for developing deep learning solution for the recognition of surgical activities in the form of a triplet. It is first surgical data science effort to replicate activity recognition in the same level as human-object interaction (HOI) in natural vision tasks.
The parent dataset is CholecT50.
Benchmarks
| Trend | Task | Dataset Variant | Best Model | Paper | Code |
|---|---|---|---|---|---|
|
CholecT40
|
Tripnet
|
Papers
| Paper | Code | Results | Date | Stars |
|---|
