Looking for Landmark (key point) estimator base model

I’ve been able to get a decent semantic segmentation for my plant dataset using unet.

However I am looking for an alternate object detection / key point estimation base model to train with my plant dataset.

FPENet appears to be face oriented custom key points, and bpnet appears to be person oriented for custom pose key points.

Before I go into one of these, and since the level of complexity and type of work to create a minimal data set for transfer learning is 3 to 5 days, I am looking for advice on which of the TAO models may be more appropriate and easy to implement.

The model should detect plants in my dataset, and estimate keypoints such as the base of the plant, the nodes where branches span out from, and the tips of leaves.

Your advise is much appreciated!!


There is fpenet-generic in next TAO release. Please stay tune.

Staying tuned…

Is there an approximate date?


About 1 or 2 months.

Topic closed as new version released: Release Notes — TAO Toolkit 3.22.05 documentation