Martyna Poreba

Graph-Guided Token Merging (G2TM) is extended, a single module inserted early in a ViT-based network, beyond its original Segmenter setting, and its behavior and accuracy-efficiency trade-off are consistent across every tested architecture for a given backbone size, indicating that its effectiveness is a property of the encoder rather than the decoder.

Achieving fully integer-only inference for lightweight detection transformers could revolutionize their deployment in resource-constrained environments.

Quantization errors in speech enhancement models can be largely mitigated by robust spatial filtering, enabling efficient deployment on low-resource devices.

Achieving ultra-low latency of 8 ms, RT-Tango redefines binaural speech enhancement for hearing aids without sacrificing performance.

Resource-strapped robots can now achieve near state-of-the-art panoptic segmentation without sacrificing accuracy, thanks to a novel lightweight architecture that slashes computational costs.