Hello DLC-Contributors,
First of all, many thanks for the great tool!
Is there any plans for the SuperAnimal model based on EfficientNet (b0)? I would be specifically interested in the SuperAnimal-Quadruped model. I have quite a performance sensitive situation, where even a 10% gain in inference speed would result in a major amount of saved computational hours. I have relatively big frames (1280x1024px), 100fps recordings from 18 cameras with duration of 1-2.5h. I have tried the ResNet-50 version of SuperAnimal-Quadruped model with transfer-learning and it performs great in my case! It basically allows to skip massive annotation efforts given multiple viewing angles, a huge variability in animal sizes due to distances to cameras, and individual differences.
It would be amazing if the power of transfer-learning would be unlocked for EfficientNet (b0) out-of-the-box too!
Best,
Vladyslav
Hello DLC-Contributors,
First of all, many thanks for the great tool!
Is there any plans for the SuperAnimal model based on EfficientNet (b0)? I would be specifically interested in the SuperAnimal-Quadruped model. I have quite a performance sensitive situation, where even a 10% gain in inference speed would result in a major amount of saved computational hours. I have relatively big frames (1280x1024px), 100fps recordings from 18 cameras with duration of 1-2.5h. I have tried the ResNet-50 version of SuperAnimal-Quadruped model with transfer-learning and it performs great in my case! It basically allows to skip massive annotation efforts given multiple viewing angles, a huge variability in animal sizes due to distances to cameras, and individual differences.
It would be amazing if the power of transfer-learning would be unlocked for EfficientNet (b0) out-of-the-box too!
Best,
Vladyslav