---
title: 'L1RA: Rotation Averaging & Rank Assignment'
url: https://www.emergentmind.com/topics/l1ra
type: topic
---

# L1RA: Rotation Averaging & Rank Assignment

Searching arXiv for the cited L1RA papers and closely related references.
L1RA is an overloaded acronym in the arXiv literature. In close-range photogrammetry, it denotes **L1 Rotation Averaging**, a method that takes as input a set of pairwise relative camera rotations and recovers a consistent set of absolute camera orientations $\{R_1,\dots,R_n\}$ before triangulation of a sparse point cloud [1702.06722]. In parameter-efficient fine-tuning of large language models, it denotes **L1-regularized Rank Assignment in LoRA Fine-Tuning**, a technique that uses $\ell_1$ regularisation on adapter gating vectors to prune redundant ranks and redistribute them across adapters under a fixed rank budget [2509.04884]. The two usages are algorithmically unrelated, but both place an $L_1$ penalty or cost at the center of the method. This suggests a shared naming logic based on robustness or sparsity rather than a shared application domain.

## 1. Polysemy and terminological scope

The term **L1RA** does not identify a single canonical method across fields. In the 2017 photogrammetry paper, the acronym is explicitly expanded as “L1 Rotation Averaging,” and its function is geometric pose recovery inside a reconstruction pipeline for temples in the Special Region of Yogyakarta [1702.06722]. In the 2025 LoRA paper, the same acronym is used for a method that dynamically distributes adapter rank during fine-tuning of quantized large language models [2509.04884].

A nearby source of ambiguity is the use of **L1** as shorthand for **first-line** treatment in clinical evidence synthesis. In the rheumatoid arthritis study on biologic therapies, “L1” refers to first-line biologic therapy effects in a bivariate network meta-analysis, not to L1RA as an algorithmic label [2201.01720]. For bibliographic and technical interpretation, the acronym therefore requires domain-specific disambiguation.

## 2. L1 Rotation Averaging as a camera-orientation estimator

In the photogrammetric usage, L1RA is introduced as the stage that follows feature matching and essential-matrix decomposition. Its input is the set of pairwise relative camera rotations, and its output is a consistent set of absolute orientations $\{R_1,\dots,R_n\}$. Once camera poses are available, 3D points can be triangulated to yield a sparse point cloud [1702.06722].

The paper does not reproduce L1RA’s objective in detail and instead cites earlier literature. A common formulation of L1 rotation averaging is

$$
\min_{R_1,\dots,R_n\,\in\,SO(3)}
\sum_{(i,j)\in \mathcal{E}}
\big\|\,R_{ij} \;-\; R_j\,R_i^\top \big\|_{1}
$$

where $R_{ij}\in SO(3)$ is the measured relative rotation from camera $i$ to $j$, $R_i\in SO(3)$ are the unknown absolute rotations, $\|\cdot\|_1$ denotes the entry-wise $L_1$ norm, and $\mathcal{E}$ is the set of image-pair edges for which a relative rotation has been estimated. A more robust variant replaces the matrix-entry $L_1$ by the geodesic $L_1$ on the Lie algebra:

$$
\min_{R_i\in SO(3)}
\sum_{(i,j)\in \mathcal{E}}
\big\|\log\bigl(R_{ij}\,R_i\,R_j^\top\bigr)\big\|_1\,,
$$

subject to each $R_i\in SO(3)$. The extracted summary explicitly notes that neither the paper nor its figures give these equations, and that they are included only to show the form of a typical L1RA problem [1702.06722].

## 3. Position within the Yogyakarta temple reconstruction pipeline

The temple-reconstruction study collected **1319 photos from five temples in Yogyakarta**. Using **A-KAZE algorithm**, keypoints of each image were obtained; **LIOP** was then employed to create feature descriptor from the detected points. After performing feature matching, **L1RA** was utilized to create sparse point clouds. To generate the geometry shape, **MVS** was used. Finally, **FSSR** and **Large Scale Texturing** were employed to deal with the surface and texture of the object [1702.06722].

At the algorithmic level, the paper describes L1RA only at a high level: build a rotation-measurement graph from matched features and pairwise relative rotations $R_{ij}$, solve a global optimization using an $L_1$ cost to mitigate the influence of outlier relative-rotation estimates, extract the final camera orientations $R_i$, and triangulate matched feature points across suitably posed cameras to produce a sparse 3D point cloud. The authors state that L1RA “significantly outperforms other methods” because the $L_1$ cost is more robust to spurious measurements. They do not report pseudocode, internal parameters, convergence tolerances, maximum iterations, or reweighting schedules; they state only that they used the implementation supplied in the **OpenMVG** library [1702.06722].

The paper also does not provide an ablation or isolated timing or accuracy study for L1RA. All concrete performance figures concern the end-to-end reconstruction pipeline and the final model quality, measured by comparing the 3D images of the model with the original photos utilizing **SSIM**. On that end-to-end criterion, the method was reported to be on par with commercial methods such as **PhotoModeler** and **PhotoScan** [1702.06722]. No temple-specific limitations of L1RA are discussed; it is treated as a drop-in replacement for other rotation-averaging schemes.

## 4. L1RA as dynamic rank assignment in LoRA fine-tuning

In the LoRA literature, L1RA addresses a different problem: standard LoRA chooses a fixed rank $r$ for every adapter, which can be sub-optimal if different layers or matrices require different capacity to adapt. The paper positions prior adaptive variants such as **ADALORA**, **SORA**, and **DYLORA** as methods that prune unused ranks but do not re-assign freed ranks to other adapters, so to reach a better allocation they must start from an inflated rank, which blows up memory [2509.04884].

L1RA instead starts with a total rank budget
$$
R_{\max}=\sum_i r_i
$$
and adds a small gating vector $c_i\in\mathbb{R}^{r_i}$ per adapter. For a pre-trained weight matrix $W_i\in\mathbb{R}^{d_{\rm in}\times d_{\rm out}}$ in layer $i$, LoRA adds
$$
A_i\in\mathbb{R}^{d_{\rm in}\times r_i},\qquad B_i\in\mathbb{R}^{r_i\times d_{\rm out}},
$$
and L1RA computes
$$
h_i(x)=x\bigl(W_i + A_i\,\mathrm{diag}(c_i)\,B_i\bigr)\,.
$$
Let $\Theta$ collect all trainable parameters $(A_i,B_i,c_i)$, and let $\mathcal{L}_{\rm task}(\Theta)$ be the downstream loss. The constrained optimization is

$$
\min_{\Theta}\;  \mathcal{L}_{\rm task}(\Theta)\;+\;\lambda\,\sum_{i=1}^N\|c_i\|_1
\qquad
\text{subject to}
\qquad
\sum_{i=1}^N\bigl|\{\,j:\,c_{i,j}\neq0\}\bigr|\;\le\;R_{\max}\,.
$$

An equivalent view is the Lagrangian
$$
\mathcal{L}(\Theta)=\mathcal{L}_{\rm task}(\Theta)+\lambda\sum_i\|c_i\|_1
$$
together with the requirement that the active rank
$$
r_i^{(\rm active)}=\#\{j:|c_{i,j}|>0\}
$$
never lets $\sum_i r_i^{(\rm active)}>R_{\max}$. The paper states that $\ell_1$ regularization is chosen because it drives individual gate components $c_{i,j}$ exactly to zero, which is the cheapest way to detect and remove a rank dimension [2509.04884].

## 5. Optimization procedure, empirical characterization, and diagnostic signals

The high-level procedure begins by choosing an initial per-adapter rank $r_i^{(0)}$ such that $\sum_i r_i^{(0)}=R_{\max}$, initializing $A_i\sim\mathcal{N}(0,\sigma^2)$, $B_i=0$, and $c_i=\mathbf{1}$. Each batch computes the forward pass using $A_i\,\mathrm{diag}(c_i)\,B_i$, optimizes $\mathcal{L}_{\rm task}+\lambda\sum_i\|c_i\|_1$, prunes every coordinate with $c_{i,j}=0$ by removing the corresponding column of $A_i$ and row of $B_i$, and then redistributes freed ranks by selecting the adapter whose minimum gate value $\min_j|c_{k,j}|$ is largest. A new column, row, and gate value are appended to that adapter, and optional normalization rescales each $c_i$ so that $\max_j|c_{i,j}|=1$ while preserving the sparsity pattern [2509.04884].

The reported fine-tuning scenario is instruction-tuning a chat-assistant on the **OPENORCA** dataset, using **MISTRAL 7B v0.3** and **LLAMA 3.1 8B**, both **4-bit quantized**. Baselines are vanilla **LoRA with rank $r=16$**, **ADALORA** starting from $r=1.5\,r$ targeting $\bar r=16$, and **ADALORA** starting at $r=16$ targeting $\bar r=12$. Metrics are **test perplexity (PPL)**, **total fine-tuning time**, **GPU memory peak**, and **\# trainable adapter parameters** [2509.04884].

| System | Configuration | Reported metrics |
|---|---|---|
| LLAMA 3.1 8B | LoRA $(r=16)$ | PPL=3.32; time≈31 k s; mem=13.84 GB; params=41.9 M |
| LLAMA 3.1 8B | ADALORA $(24\to16)$ | PPL≈3.63 (+9%); time≈32.98 k s (+6%); mem≈14.00 GB (+1%); params≈62.9 M (+50%) |
| LLAMA 3.1 8B | ADALORA $(16\to12)$ | PPL≈3.57 (+7.5%); time≈32.96 k s (+6.3%); mem≈13.76 GB (−0.6%); params≈42.0 M (+0.1%) |
| LLAMA 3.1 8B | **L1RA** $(r=16)$ | **PPL=3.25** (−2.1%); time=31.25 k s (+0.8%); mem=14.23 GB (+2.8%); params=45.2 M (+7.7%) |
| MISTRAL 7B v0.3 | LoRA | PPL=2.93; time=37.9 k s; mem=13.58 GB; params=41.9 M |
| MISTRAL 7B v0.3 | ADALORA $(24\to16)$ | PPL=3.16 (+7.9%); time=40.23 k s (+6.2%); mem=13.82 GB (+1.8%); params=62.9 M (+50%) |
| MISTRAL 7B v0.3 | ADALORA $(16\to12)$ | PPL=3.16 (+7.9%); time=40.22 k s (+6.1%); mem=13.59 GB (+0.1%); params=42.0 M (+0.1%) |
| MISTRAL 7B v0.3 | **L1RA** $(r=16)$ | **PPL=2.91** (−0.7%); time=37.97 k s (+0.2%); mem=13.94 GB (+2.6%); params=50.1 M (+19.4%) |

The paper’s stated takeaway is that L1RA matches or improves over vanilla LoRA in perplexity while keeping within the original memory footprint, avoids the large parameter-count blow-up of ADALORA, and incurs only a few percent of extra VRAM. Its post-training rank-distribution analysis reports that the highest ranks are consistently assigned to the three sub-matrices of the Transformer feed-forward network—$W_{\rm up}$, $W_{\rm gate}$, and $W_{\rm down}$—while the attention matrices $W_q$, $W_k$, $W_v$, and $W_o$ receive lower ranks, and ranks tend to increase in layers closer to the final output. The paper states that this suggests that adapting high-level FFN transforms is more critical for instruction-tuning than reshaping self-attention at every layer [2509.04884].

## 6. Comparative interpretation, limitations, and recurrent misconceptions

The two meanings of L1RA should not be conflated. In photogrammetry, the method estimates elements of $SO(3)$ from a rotation-measurement graph and enables triangulation of sparse 3D structure before later stages such as MVS, FSSR, and Large Scale Texturing [1702.06722]. In LoRA fine-tuning, the method operates on trainable low-rank adapters $(A_i,B_i,c_i)$ and reallocates active rank under a global budget $R_{\max}$ [2509.04884]. The commonality is methodological emphasis on an $L_1$ criterion: outlier resistance in one case, exact sparsification of gate coordinates in the other.

A common misconception is to read **L1RA** as a stable cross-domain acronym. The rheumatoid arthritis paper illustrates why this is unreliable: there, **L1** denotes **first-line** therapy, with covariate-adjusted logistic regression estimates $\hat\theta_{1i}$ for first-line contrasts entering a bivariate network meta-analysis, but no L1RA algorithm is involved [2201.01720]. Another misconception is to overgeneralize reported performance. The temple paper provides no ablation or isolated timing or accuracy study of L1RA itself, and its concrete evaluation concerns the end-to-end reconstruction pipeline [1702.06722]. The LoRA paper reports validation on **two 8 billion-parameter models** and **a single instruction-tuning task**, and explicitly identifies further benchmarks across scale, tasks, and architectures as a valuable next step [2509.04884].

Taken together, the available literature supports a narrow and technically specific reading of each usage. **L1 Rotation Averaging** is a globally optimal, outlier-robust pose-estimation component in close-range photogrammetry, whereas **L1-regularized Rank Assignment** is a budget-constrained adapter-allocation strategy for efficient LLM fine-tuning. Any interpretation of “L1RA” therefore depends primarily on the surrounding research domain, the objects being optimized, and the role of the $L_1$ term in the objective.

Source: https://www.emergentmind.com/topics/l1ra