--- title: Unsupervised Exposure Correction (UEC) url: https://www.emergentmind.com/topics/unsupervised-exposure-correction-uec type: topic --- # Unsupervised Exposure Correction (UEC) Searching arXiv for the specified paper and closely related unsupervised exposure correction work to ground the article in current literature. arXiv search query: "2511.06066 LoopExpose unsupervised exposure correction PSENet 2507.17252 UNICE 2601.19129" Unsupervised exposure correction (UEC) denotes exposure-correction methods that avoid human-annotated well-exposed targets while still seeking to restore detail, contrast, and radiometric plausibility in under-exposed, over-exposed, or mixed-exposure imagery. In recent arXiv literature, UEC has been instantiated through pseudo label refinement from multi-exposure fusion, progressive pseudo-ground-truth synthesis, RAW-to-sRGB ISP emulation, closed-form iterative compensation, HDR-derived pseudo supervision, and semantic-prompt guidance from foundation models [2511.06066; 2210.00712; 2507.17252; 2212.14245; 2507.17157; 2601.19129]. The term therefore refers both to a research area and, in one case, to a specific 2025 method titled "Unsupervised Exposure Correction" [2507.17252]. ## 1. Problem setting and scope Exposure correction is framed in the cited works as the restoration of visually usable images from inputs whose radiometry is too low, too high, or spatially imbalanced. "LoopExpose" formulates the task for an arbitrary-length sequence of under/over-exposed RGB images, whereas UNICE formulates it as mapping a single 8-bit sRGB image $I_{in}$ to an enhanced 8-bit sRGB image $I_{out}$ through pseudo multi-exposure sequence generation and fusion [2511.06066; 2507.17157]. "PSENet" addresses "extreme-light image enhancement," explicitly targeting both under-exposure and over-exposure rather than only low-light cases [2210.00712]. "CLIP-Guided Unsupervised Semantic-Aware Exposure Correction" adds the claim that improper exposure often causes severe loss of details, color distortion, and reduced contrast, and attributes color-shift artifacts partly to the ignorance of object-wise regional semantic information [2601.19129]. The principal motivation across these works is the impracticality of paired supervision. "LoopExpose" states that supervised learning has achieved significant progress but relies heavily on large-scale labeled datasets that are difficult to obtain in practical scenarios [2511.06066]. The 2025 "Unsupervised Exposure Correction" paper is more specific: expert-retouched targets are labor-intensive, introduce individual style biases, and can reduce cross-dataset generalization; large image-to-image models may also degrade low-level features such as edges and fine textures [2507.17252]. "Practical Exposure Correction: Great Truths Are Always Simple" takes an even stronger position by removing learning entirely and recasting exposure correction as estimation of a single exposure-sensitive compensation map $\bm{\epsilon}(\mathbf y)$ in a linear add/subtract model [2212.14245]. A useful distinction within UEC is between single-image and multi-exposure settings. LoopExpose operates on exposure sequences and explicitly couples single-exposure correction (SEC) with multi-exposure fusion (MEF), while PSENet, PEC, the 2025 UEC model, UNICE, and the CLIP-guided method all begin from a single sRGB image and construct additional supervisory structure internally [2511.06066; 2210.00712; 2212.14245; 2507.17252; 2507.17157; 2601.19129]. This suggests that contemporary UEC is less defined by input cardinality than by the rejection of manually retouched ground truth. ## 2. Pseudo supervision and target construction A defining property of UEC is the replacement of human-retouched references with automatically generated targets. The mechanisms differ substantially. LoopExpose uses a classical Mertens multi-exposure fusion operator $\mathcal M$ to construct pseudo-labels. In warm-up, the initial pseudo-label is produced only from raw exposures, $Y^{(0)}=\mathcal M(I)$; during joint optimization, the corrected outputs are fed back into fusion so that $$ Y^{(t+1)}=\mathcal M\bigl(I_1^{obs},\ldots,I_N^{obs},E_1^{(t)},\ldots,E_N^{(t)}\bigr), $$ creating the paper’s self-reinforcing loop in which better $E$ yields better $Y$ and vice versa [2511.06066]. PSENet constructs a bank of darker and brighter reference images from a single input by invert-gamma mapping, then scores each candidate using non-reference measures of well-exposedness, local contrast, and color saturation. Its pseudo-ground-truth is obtained by pixel-wise selection: $$ Q_n(x,y)=\frac{C_n(x,y)S_n(x,y)}{E_n(x,y)+\epsilon}, \qquad T(x,y)=Y_{n^*}(x,y), \quad n^*=\arg\max_n Q_n(x,y). $$ Because the candidate set includes the previous-epoch output $\hat Y_{prev}$, the pseudo-GT can only improve or stay equal from epoch to epoch, which motivates the term Progressive Self-Enhancement [2210.00712]. The 2025 UEC model dispenses with pseudo-label synthesis from the observed image itself and instead exploits an emulated ISP pipeline. Starting from RAW, it applies calibrated gain for $\Delta EV\in\{-2,-1,0,+1,+2,+3\}$, followed by demosaicing, white balance, color-space conversion, and gamma correction, producing exposure brackets with known scalar EV shifts. Training then uses image pairs or triplets sampled from these brackets as pseudo-paired supervision [2507.17252]. UNICE scales this RAW-based idea to 46,928 HDR raw images rendered into 328,496 sRGB images across $EV\in\{-3,-2,-1,0,+1,+2,+3\}$. It forms pseudo sRGB ground-truths by applying several classical/fusion algorithms to MES triplets, ensembling them by NR-IQA ranking, and filtering out poor results with $\text{ARNIQA}<0.5$, leaving 328,496 input-GT pairs [2507.17157]. The CLIP-guided semantic-aware method replaces handcrafted quality scores and physical brackets with prompt-conditioned pseudo-GT generation. It fine-tunes three text prompts for well-, under-, and over-exposed images, compares CLIP similarities $sim_u$ and $sim_o$, and then applies gamma-brightening or gamma-darkening accordingly. It can further refine $\gamma$ by maximizing similarity to the well-exposed prompt in CLIP’s joint space [2601.19129]. These strategies can be summarized as follows. | Method | Pseudo-supervision source | Core mechanism | |---|---|---| | LoopExpose [2511.06066] | Mertens MEF pseudo-labels | Nested feedback loop | | PSENet [2210.00712] | Progressive pseudo-GT bank | Pixel-wise quality selection | | PEC [2212.14245] | None beyond input image | Closed-form compensation | | UEC [2507.17252] | Emulated ISP exposure brackets | Known EV-shift pseudo-pairs | | UNICE [2507.17157] | HDR-rendered MES + MEF pseudo-GTs | Two-stage MES generation and fusion | | Semantic-aware method [2601.19129] | CLIP-guided pseudo-GT | Prompt-conditioned gamma tuning | ## 3. Model families and optimization regimes LoopExpose is organized as a bi-level optimization. The upper level updates a correction model with parameters $\theta$ by minimizing a supervised loss against pseudo-labels together with a luminance constraint; the lower level refines pseudo-labels with fixed rule-based fusion. Training is divided into warm-up, with decayed learning rate and static $Y^{(0)}=\mathcal M(I)$, and joint optimization, with constant learning rate and dynamic pseudo-label updates [2511.06066]. Architecturally, it uses a Luminance-Aware Network in encoder-decoder form, an Adaptive 3D-LUT Module, and attention-based fusion of luminance and color paths [2511.06066]. PSENet uses a lightweight U-Net-style encoder-decoder that predicts a per-pixel gamma map $\gamma\in\mathbb R^{H\times W\times 3}$ and reconstructs the enhanced image by $$ \hat Y = 1-(1-I)^\gamma. $$ Its encoder uses MobileNet-V2 blocks, while the decoder is symmetric with skip-connections and ends with a MobileNet-V3 layer to regress three gamma channels [2210.00712]. PEC is structurally distinct because it is zero-training and zero-learning. Its central adversarial generator is $$ f(\mathbf z)=c\,[\mathbf z\otimes(1-\mathbf z)], \qquad 0\le c\le 1, $$ which is axisymmetric and mid-tone emphasizing. For under-exposure, it iterates $$ \mathbf x_u^{(k)} = g + f\bigl(\mathbf x_u^{(k-1)}\bigr), \qquad g=\mathbf y+f(\mathbf y), $$ within a segmented shrinkage scheme of up to $T$ built-in blocks and per-block iteration counts $K^t$ [2212.14245]. The same machinery is used for over-exposure by changing the sign. The 2025 paper titled "Unsupervised Exposure Correction" introduces a small learned radiometric corrector. Given an exposure-difference estimate $\Delta E$, its transformation block performs a pixelwise interpolation between linear scaling and a learned nonlinear curve: $$ I_{out}(x,y)=\lambda\cdot I_{in}(x,y)+(1-\lambda)\cdot h(I_{in}(x,y)), $$ where $\lambda\in[0,1]$ is predicted from $\Delta E$ by a small MLP and $h(\cdot)$ is implemented by three sequential $1\times1$ convolutions with ReLU. The block is iterated three times. The full system comprises an Exposure Feature Encoder, a Difference Predictor, and an Exposure Corrector, for a total of 19,388 parameters [2507.17252]. UNICE adopts a substantially larger two-stage architecture. MES-Net maps a single image to a pseudo multi-exposure sequence using a pretrained SD-Turbo diffusion model with frozen original weights and inserted LoRA adapters; MEF-Net then predicts pixelwise weight maps, forms an implicit HDR representation $$ H(x,y)=\sum_{j=1}^{T} w'_j(x,y)\,I_j(x,y), $$ and refines it with another one-step diffusion model to produce the final enhanced output [2507.17157]. The CLIP-guided semantic-aware model is an encoder-decoder built from Semantics-Informed Mamba Reconstruction blocks. Each block first applies Adaptive Semantic-Aware Fusion, which injects FastSAM semantic features into the image feature space, then a Residual Spatial Mamba Group, which augments a Vision Mamba Module with spatial attention [2601.19129]. This architecture directly addresses the claim that object-wise semantics matter for exposure correction under mixed lighting. ## 4. Objective functions, priors, and self-supervised constraints The losses used in UEC are not uniform; they encode different hypotheses about what must be preserved when explicit ground truth is absent. LoopExpose defines $$ L_{total}=L_{sup}+L_{lumi}, $$ with $$ L_{sup}=\|E_i-Y_i\|_1 + 0.1\cdot L_{perc}(E_i,Y_i) + 0.05\cdot L_{SSIM}(E_i,Y_i). $$ Its distinctive self-supervised term is the Luminance Ranking Loss $$ L_{lumi}=w_{lumi}\sum_{i