[KongchangAI]
ModelSurgical Graph Transformer

SGT

A baseline method for surgical scene captioning that models the surgical scene as a graph and performs relational reasoning with a Transformer architecture, passing information among instrument, action, and anatomical structure nodes via graph attention mechanisms.

Timeline (last 90 days)

Nov 9

SGT refers to Surgical Graph Transformer, a baseline method that models the surgical scene as a graph and uses the Transformer architecture for relational reasoning, passing information among instrument, action and anatomical structure nodes via a graph attention mechanism.

Unverified50%
Nov 9

SGT mainly relies on knowledge stored in model parameters for generation and does not introduce an external retrieval mechanism to provide traceable reference examples.

Unverified50%
Nov 9

On the EndoVis-18 benchmark, GRAC achieves a CIDEr-D score of 5.88, SGT achieves 4.23, and the strongest traditional Transformer baseline achieves 3.36.

Unverified50%
Nov 9

On the factual consistency metric of the EndoVis-18 benchmark, GRAC's instrument-action-target triplet set accuracy reaches 52.6%, SGT reaches 32.7%, and the strongest traditional baseline reaches 26.2%.

Unverified50%

All Facts (4)

Source Articles