Lite3R cuts latency by 1.7-2.0x and memory by 1.9-2.4x in feed-forward 3D reconstruction using sparse linear attention and FP8-aware quantization-aware training while keeping competitive quality on backbones like VGGT and DA3-Large.
Train or fine-tune the model with its original precision (typically FP32, FP16, or BF16) until convergence
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Lite3R: A Model-Agnostic Framework for Efficient Feed-Forward 3D Reconstruction
Lite3R cuts latency by 1.7-2.0x and memory by 1.9-2.4x in feed-forward 3D reconstruction using sparse linear attention and FP8-aware quantization-aware training while keeping competitive quality on backbones like VGGT and DA3-Large.