ARQ-VAE for Memory-Efficient Temporal Action Segmentation

The ARQ-VAE method uses adaptive residual quantization to create compact, label-free memory representations for video features. It adaptively truncates residual chains on a per-frame basis to reduce the storage bottleneck in long-form video tasks and mitigate catastrophic forgetting.