Recent advances in generalizable Gaussian splatting (GS) have enabled feed-forward reconstruction of scenes
from tens of input views. Long-LRM notably scales this paradigm to 32 input images at 960×540 resolution,
achieving 360° scene-level reconstruction in a single forward pass. However, directly predicting millions of
Gaussian parameters at once remains highly error-sensitive: small inaccuracies in positions or other
attributes lead to noticeable blurring, particularly in fine structures such as text.
In parallel, implicit representation methods such as LVSM and LaCT have demonstrated significantly higher
rendering fidelity by compressing scene information into model weights rather than explicit Gaussians,
and decoding RGB frames using the full transformer or TTT backbone.
However, this computationally intensive decompression process for every rendered frame makes real-time
rendering infeasible.
These observations raise key questions: Is the deep, sequential "decompression" process necessary?
Can we retain the benefits of implicit representations while enabling real-time performance?
We address these questions with Long-LRM++, a model that adopts a semi-explicit scene representation
combined with a lightweight decoder. Long-LRM++ matches the rendering quality of LaCT on DL3DV while
achieving real-time 14 FPS rendering on an A100 GPU, overcoming the speed limitations of prior
implicit methods. Our design also scales to 64 input views at the 950×540 resolution, demonstrating
strong generalization to increased input lengths.
Additionally, Long-LRM++ delivers superior novel-view depth prediction on ScanNetv2 compared to direct
depth rendering from Gaussians. Extensive ablation studies validate the effectiveness of each component
in the proposed framework.
Long-LRM++ takes up to 64 input images at 950×540 resolution along with their camera poses,
and processes them using a backbone composed of interleaved Mamba2 and Transformer blocks,
similar to Long-LRM. Each image token predicts $K$ free-moving feature Gaussians
(visualized with originating pixel's color). During rendering, we introduce a multi-space partitioning
step that divides the Gaussians into multiple subsets, each rendered and decoded independently.
The target-frame decoder incorporates translation-invariant local-attention blocks to improve robustness
and rendering quality. Finally, the decoded feature maps are merged and passed through a linear layer to
produce the novel-view color or depth rendering.