VGG-T$^3$: Offline Feed-Forward 3D Reconstruction at Scale
This work provides a significant speed-up for researchers and practitioners working on large-scale offline 3D reconstruction, making it more feasible to process extensive image collections.
The paper introduces VGG-T^3, a scalable 3D reconstruction model that overcomes the quadratic scaling of offline feed-forward methods by distilling the varying-length Key-Value space into a fixed-size MLP. This approach enables linear scaling with the number of input views, reconstructing a 1k image collection in 54 seconds, which is an 11.6x speed-up over softmax attention baselines.
We present a scalable 3D reconstruction model that addresses a critical limitation in offline feed-forward methods: their computational and memory requirements grow quadratically w.r.t. the number of input images. Our approach is built on the key insight that this bottleneck stems from the varying-length Key-Value (KV) space representation of scene geometry, which we distill into a fixed-size Multi-Layer Perceptron (MLP) via test-time training. VGG-T$^3$ (Visual Geometry Grounded Test Time Training) scales linearly w.r.t. the number of input views, similar to online models, and reconstructs a $1k$ image collection in just $54$ seconds, achieving a $11.6\times$ speed-up over baselines that rely on softmax attention. Since our method retains global scene aggregation capability, our point map reconstruction error outperforming other linear-time methods by large margins. Finally, we demonstrate visual localization capabilities of our model by querying the scene representation with unseen images.