CurveCodec 2: Skeleton-agnostic animation compression with a learned entropy model
CurveCodec 2 compresses skeletal animation to 0.22–0.37× ACL's size with a learned residual entropy model.
CurveCodec 2 encodes skeletal motion as log-map curves, quantized in closed loop and thinned to rate-distortion keys, then entropy-codes residuals with a small learned model whose integer inference is bit-exact across platforms. On 4,472 held-out clips from 33 datasets it needs 0.37× ACL's bytes at ACL's 0.01 cm worst-case precision and 0.22× at 0.1 cm under the mean-error contract. It decodes on one CPU core and transfers without retraining to a species absent from training. Learned in-betweeners did not beat linear interpolation on the gaps the encoder leaves.
- Uses 0.37× ACL bytes at 0.01 cm under the worst-case contract
- Uses 0.22× ACL bytes at 0.1 cm under the mean-error contract
- Learned model entropy-codes residuals with bit-exact integer inference
- Transfers to a species absent from training without retraining
- Evaluated on 4,472 held-out clips from 33 datasets
- Products
- CurveCodec 2ACL
From the original article
1 min read at huggingface.co
Skeletal motion is stored as every joint's transform at every frame, yet most of it is implied by the body rather than by what the motion is about. Compression is one way to ask what a motion must still say once the body is known, and a production codec must answer it for any skeleton with a stated error bound.
By Mingyi Shi, Huancheng Lin, Xuelin Chen · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.