Caleb posted that AMW doesn't use embeddings and instead an ensemble of CNNs.
Their repo in fact both hosts directly the AoIs, labels, and [dissolved] outputs.
I feel it should not be hard to both benchmark detection with the same input data and compare outputs, but also finetune Clay with either/both inputs and outputs.
brunosan
commented
2 months ago
Caleb posted that AMW doesn't use embeddings and instead an ensemble of CNNs.
Their repo in fact both hosts directly the AoIs, labels, and [dissolved] outputs.
I feel it should not be hard to both benchmark detection with the same input data and compare outputs, but also finetune Clay with either/both inputs and outputs.