Diffusion Contrastive Reconstruction (DCR)

• Diffusion Contrastive Reconstruction (DCR) is a method that fuses diffusion-based reconstruction with contrastive guidance to boost both detail fidelity and class separability.
• In vision, DCR addresses the trade-off between perceptual detail and discriminative geometry, mitigating gradient conflicts and improving CLIP model performance.
• In audio deepfake detection, DCR generates hard samples via diffusion and employs Regularization-Assisted Contrastive Learning to enhance robustness against unseen attacks.

Diffusion Contrastive Reconstruction (DCR) denotes a class of methods that combine diffusion-based reconstruction with contrastive supervision, but the exact meaning is domain-dependent. In visual representation learning, DCR is a unified objective that injects contrastive signals into diffusion reconstruction so that CLIP visual encoders improve both Discriminative Ability (D-Ability) and Detail Perceptual Ability (P-Ability) (Han et al., 5 Mar 2026). In audio deepfake detection, the same expansion refers to a framework that uses diffusion-based reconstruction to generate hard samples and trains a detector with Regularization-Assisted Contrastive Learning (RACL) for improved generalization to unseen attacks (Cheng et al., 29 Apr 2026). The acronym also appears in unrelated contexts, including low-light denoising, synthetic-data privacy evaluation, and rare compositional generation, making contextual disambiguation necessary (Cui et al., 2023, Yao et al., 2 May 2025, Kang et al., 7 May 2026).

1. Terminological scope and disambiguation

In the cited literature, DCR refers to multiple distinct constructs rather than a single standardized method.

Usage of DCR	Domain	Defining paper
Diffusion Contrastive Reconstruction	Balanced visual representation from CLIP-guided diffusion reconstruction	(Han et al., 5 Mar 2026)
Diffusion Contrastive Reconstruction	Generalizable audio deepfake detection via hard-sample reconstruction	(Cheng et al., 29 Apr 2026)
Denoising Contrastive Regularization	Low-light RAW denoising	(Cui et al., 2023)
Distance to Closest Record	Synthetic-data privacy proxy metric	(Yao et al., 2 May 2025)
Default Completion Repulsion	Rare compositional diffusion generation	(Kang et al., 7 May 2026)

Within the diffusion-and-representation literature, the most direct use of Diffusion Contrastive Reconstruction is the visual-representation formulation of Hu et al., which addresses the tension between discriminative geometry and perceptual detail in CLIP encoders (Han et al., 5 Mar 2026). A second explicit use of the same expansion appears in audio deepfake detection, where diffusion reconstruction is used to synthesize hard samples rather than to enhance image embeddings (Cheng et al., 29 Apr 2026). These two uses share the coupling of reconstruction and contrastive structure, but they operate on different modalities, objectives, and evaluation protocols.

2. Representation-learning formulation in vision

The visual DCR formulation begins from the claim that the “understanding capacity” of CLIP visual encoders is limited in two complementary ways: D-Ability, which concerns class separability, and P-Ability, which concerns fine-grained visual detail such as color, count, spatial relation, viewpoint, and structure (Han et al., 5 Mar 2026). Prior diffusion-based enhancement methods, including DIVA, GenHancer, and un$^2$CLIP, improve reconstruction-oriented perceptual detail but may not improve, and can even hurt, discriminative structure.

The paper formalizes D-Ability with intra-class and inter-class scatter. For class centers $\mu_y$,

$$S_{\text{inner}}=\frac{1}{|\mathcal Y|}\sum_{y\in\mathcal Y}\frac{1}{|\mathcal Z_y|}\sum_{\mathbf z\in\mathcal Z_y}\|\mathbf z-\mu_y\|_2^2,$$

$$S_{\text{inter}}=\frac{1}{|\mathcal Y|(|\mathcal Y|-1)}\sum_{y\neq y'}\|\mu_y-\mu_{y'}\|_2^2.$$

The desired regime is

$$\min S_{\text{inner}}, \qquad \max S_{\text{inter}}.$$

The paper formulates P-Ability as reconstruction consistency under diffusion decoding: $$\min \mathbb{E}_{t}\Big[\|\epsilon_{\theta}(\tilde{x}_t,\ h_{\omega}(f_{\phi}(x)),t)-\epsilon_t^{\text{gt}}\|_2^2\Big].$$ Here, $f_\phi$ is the CLIP visual encoder, $h_\omega$ maps CLIP features into the diffusion conditioning space, and $\epsilon_\theta$ is the diffusion denoiser.

The central problem is that these two desiderata are not automatically aligned. A representation optimized only for reconstruction can become rich in local detail while losing class geometry; a representation optimized only with contrastive learning can improve separability while failing to preserve the information needed for faithful reconstruction. DCR is introduced precisely to avoid that tradeoff by integrating contrastive supervision into the diffusion reconstruction process itself rather than summing two external objectives (Han et al., 5 Mar 2026).

3. Unified objective and noise-space contrastive learning

The paper first studies a naïve multi-objective formulation,

$$\mathcal{L}_{\text{joint}}=\lambda_{\text{con}}\mathcal{L}_{\text{con}}+\lambda_{\text{rec}}\mathcal{L}_{\text{rec}},$$

and attributes its weakness to gradient conflict between feature-space discrimination and image-level reconstruction (Han et al., 5 Mar 2026). The conflict is quantified by gradient cosine similarity,

$\mu_y$0

with 86.3% of training steps showing negative cosine similarity. The paper further states that the conflict intensifies during training and causes the contrastive term to dominate while reconstruction stalls.

DCR replaces this two-loss design with a single objective defined in predicted-noise space. The anchor is the predicted noise conditioned on the original image,

$\mu_y$1

A positive is obtained from an augmented view $\mu_y$2,

$\mu_y$3

and a second positive is the ground-truth diffusion noise $\mu_y$4. Negatives are predicted noises from other images in the mini-batch. With

$\mu_y$5

$\mu_y$6

the DCR loss is

$\mu_y$7

This formulation makes the diffusion model’s predicted noise the common space in which reconstruction and contrastive structure meet. The paper’s theoretical analysis states that, under a mild bi-Lipschitz assumption on

$\mu_y$8

the feature geometry inherits noise-space improvements: $\mu_y$9

$$S_{\text{inner}}=\frac{1}{|\mathcal Y|}\sum_{y\in\mathcal Y}\frac{1}{|\mathcal Z_y|}\sum_{\mathbf z\in\mathcal Z_y}\|\mathbf z-\mu_y\|_2^2,$$0

Under separation assumptions, the paper further states that DCR is equivalent, up to scaling and constants, to the reconstruction loss

$$S_{\text{inner}}=\frac{1}{|\mathcal Y|}\sum_{y\in\mathcal Y}\frac{1}{|\mathcal Z_y|}\sum_{\mathbf z\in\mathcal Z_y}\|\mathbf z-\mu_y\|_2^2,$$1

The intended implication is that DCR jointly supports both discriminative structure and reconstruction fidelity without explicitly maintaining two conflicting losses (Han et al., 5 Mar 2026).

4. Architecture, training protocol, and empirical behavior in vision

The visual DCR system uses Stable Diffusion v2.1 as the main diffusion backbone, a 2-layer MLP projector, and the CLIP [CLS] token only as condition (Han et al., 5 Mar 2026). Training data is CC3M. Optimization uses AdamW, weight decay 0.01, batch size 16, and 4600 total steps. The training protocol is explicitly two-stage: in Stage 1 the CLIP encoder $$S_{\text{inner}}=\frac{1}{|\mathcal Y|}\sum_{y\in\mathcal Y}\frac{1}{|\mathcal Z_y|}\sum_{\mathbf z\in\mathcal Z_y}\|\mathbf z-\mu_y\|_2^2,$$2 is frozen and the projector $$S_{\text{inner}}=\frac{1}{|\mathcal Y|}\sum_{y\in\mathcal Y}\frac{1}{|\mathcal Z_y|}\sum_{\mathbf z\in\mathcal Z_y}\|\mathbf z-\mu_y\|_2^2,$$3 is trained with learning rate $$S_{\text{inner}}=\frac{1}{|\mathcal Y|}\sum_{y\in\mathcal Y}\frac{1}{|\mathcal Z_y|}\sum_{\mathbf z\in\mathcal Z_y}\|\mathbf z-\mu_y\|_2^2,$$4; in Stage 2 the projector is frozen, the CLIP encoder is refined with DCR loss at learning rate $$S_{\text{inner}}=\frac{1}{|\mathcal Y|}\sum_{y\in\mathcal Y}\frac{1}{|\mathcal Z_y|}\sum_{\mathbf z\in\mathcal Z_y}\|\mathbf z-\mu_y\|_2^2,$$5, and LoRA rank 16 is applied to the vision encoder.

The method is evaluated on six CLIP backbones: OpenAI CLIP ViT-L@224, OpenAI CLIP ViT-L@336, MetaCLIP ViT-L@224, MetaCLIP ViT-H@224, SigLIP ViT-SO@224, and SigLIP ViT-SO@384. P-Ability is evaluated on MMVP-VLM; D-Ability is evaluated by zero-shot clustering on MNIST, CIFAR-10, EuroSAT, Caltech-101, DTD, and ImageNet-1K, using NMI, ACC, and ARI; transfer is tested by inserting the enhanced CLIP into LLaVA-1.5 and evaluating on MMVP-MLLM, NaturalBench, CV-Bench 2D, CV-Bench 3D, POPE, SciQA-IMG, and HallusionBench (Han et al., 5 Mar 2026).

Quantitatively, the paper reports that on MMVP-VLM DCR improves OpenAI CLIP ViT-L@224 by +14.1% and MetaCLIP ViT-L@224 by +8.9%, while consistently outperforming DIVA, GenHancer, and un$$S_{\text{inner}}=\frac{1}{|\mathcal Y|}\sum_{y\in\mathcal Y}\frac{1}{|\mathcal Z_y|}\sum_{\mathbf z\in\mathcal Z_y}\|\mathbf z-\mu_y\|_2^2,$$6CLIP. On zero-shot clustering, SigLIP ViT-SO@224 reaches NMI / ACC / ARI = 0.83 / 0.76 / 0.65. When inserted into LLaVA-1.5 with Vicuna-7B, the enhanced CLIP improves MMVP-MLLM to 31.3, while NaturalBench and CV-Bench also improve.

The ablation results are central to the method’s interpretation. The naïve linear combination improves D-Ability somewhat but degrades P-Ability severely, reaching only 22.96% on MMVP-VLM, whereas DCR reaches 33.30% and also improves clustering metrics. The paper states that two-stage training is consistently better than end-to-end training. It also reports that stronger diffusion backbones generally improve results, that SD-2.1 performs best overall, and that SD-XL is not fully exploited because its dual-encoder conditioning does not perfectly match the single image embedding setup. A further ablation shows that using more local tokens degrades performance because reconstruction becomes too easy, weakening the supervision signal (Han et al., 5 Mar 2026).

5. Diffusion Contrastive Reconstruction in audio deepfake detection

In audio deepfake detection, Diffusion Contrastive Reconstruction designates a different framework whose goal is robustness to unseen attacks rather than improvement of visual embeddings (Cheng et al., 29 Apr 2026). The core claim is that a detector that can distinguish hard samples that are extremely close to real speech should also handle easier fake samples effectively. The method therefore reconstructs both bona fide and spoof utterances to create hard training cases that preserve content while introducing subtle generative artifacts.

The paper evaluates four reconstruction paradigms: HiFi-GAN, DAC, Encodec, and the diffusion-based SemantiCodec. The best-performing paradigm is the diffusion-based one implemented with SemantiCodec. The detection model uses a frozen XLS-R 300M pretrained self-supervised speech model and a trainable AASIST classifier. Rather than using only the top transformer layer, the method performs multi-layer feature aggregation. If the $$S_{\text{inner}}=\frac{1}{|\mathcal Y|}\sum_{y\in\mathcal Y}\frac{1}{|\mathcal Z_y|}\sum_{\mathbf z\in\mathcal Z_y}\|\mathbf z-\mu_y\|_2^2,$$7-th XLS-R output is

$$S_{\text{inner}}=\frac{1}{|\mathcal Y|}\sum_{y\in\mathcal Y}\frac{1}{|\mathcal Z_y|}\sum_{\mathbf z\in\mathcal Z_y}\|\mathbf z-\mu_y\|_2^2,$$8

the per-layer descriptor is

$$S_{\text{inner}}=\frac{1}{|\mathcal Y|}\sum_{y\in\mathcal Y}\frac{1}{|\mathcal Z_y|}\sum_{\mathbf z\in\mathcal Z_y}\|\mathbf z-\mu_y\|_2^2,$$9

the attention weights are

$$S_{\text{inter}}=\frac{1}{|\mathcal Y|(|\mathcal Y|-1)}\sum_{y\neq y'}\|\mu_y-\mu_{y'}\|_2^2.$$0

and the aggregated feature is

$$S_{\text{inter}}=\frac{1}{|\mathcal Y|(|\mathcal Y|-1)}\sum_{y\neq y'}\|\mu_y-\mu_{y'}\|_2^2.$$1

Training is based on RACL, which combines classification, standard contrastive separation, hard-sample-focused contrastive separation, and variance regularization: $$S_{\text{inter}}=\frac{1}{|\mathcal Y|(|\mathcal Y|-1)}\sum_{y\neq y'}\|\mu_y-\mu_{y'}\|_2^2.$$2

$$S_{\text{inter}}=\frac{1}{|\mathcal Y|(|\mathcal Y|-1)}\sum_{y\neq y'}\|\mu_y-\mu_{y'}\|_2^2.$$3

with

$$S_{\text{inter}}=\frac{1}{|\mathcal Y|(|\mathcal Y|-1)}\sum_{y\neq y'}\|\mu_y-\mu_{y'}\|_2^2.$$4

The enhanced contrastive term focuses specifically on bona fide versus reconstructed bona fide, which the paper treats as the hardest case because reconstructed bona fide speech is close to real speech while still containing reconstruction artifacts. The regularization term enforces batch-level intra-class compactness: $$S_{\text{inter}}=\frac{1}{|\mathcal Y|(|\mathcal Y|-1)}\sum_{y\neq y'}\|\mu_y-\mu_{y'}\|_2^2.$$5

Evaluation uses Equal Error Rate (EER) on ASVspoof 2019 LA eval, CodecFake, DiffSSD, WaveFake, and ITW. The reported average EERs over the five test sets are 15.789 for the baseline, 20.862 for HiFi-GAN, 21.922 for DAC, 14.247 for Encodec, 12.220 for diffusion reconstruction, 8.888 for Agg Diffusion, and 8.247 for RACL Diffusion. The paper explicitly states a 22.604% relative reduction in average EER compared to the baseline for diffusion reconstruction alone: $$S_{\text{inter}}=\frac{1}{|\mathcal Y|(|\mathcal Y|-1)}\sum_{y\neq y'}\|\mu_y-\mu_{y'}\|_2^2.$$6 An ablation shows 10.328 for $$S_{\text{inter}}=\frac{1}{|\mathcal Y|(|\mathcal Y|-1)}\sum_{y\neq y'}\|\mu_y-\mu_{y'}\|_2^2.$$7 only, 8.888 after adding $$S_{\text{inter}}=\frac{1}{|\mathcal Y|(|\mathcal Y|-1)}\sum_{y\neq y'}\|\mu_y-\mu_{y'}\|_2^2.$$8, 8.640 after adding $$S_{\text{inter}}=\frac{1}{|\mathcal Y|(|\mathcal Y|-1)}\sum_{y\neq y'}\|\mu_y-\mu_{y'}\|_2^2.$$9, and 8.247 after adding $$\min S_{\text{inner}}, \qquad \max S_{\text{inter}}.$$0. The t-SNE analysis further reports that $$\min S_{\text{inner}}, \qquad \max S_{\text{inter}}.$$1 increases the average distance between bona fide and other categories from 65.39 to 71.84, that separation between bona fide and reconstructed bona fide becomes 75.72 compared to a baseline distance of 50.64, and that adding $$\min S_{\text{inner}}, \qquad \max S_{\text{inter}}.$$2 increases the overall distance further to 76.95 while tightening intra-class clusters (Cheng et al., 29 Apr 2026).

6. Relation to adjacent contrastive–diffusion research

Several neighboring lines of work help situate DCR within a broader family of contrastive–reconstruction methods, while also showing that the acronym is not unique to diffusion contrastive reconstruction.

In low-light restoration, Denoising Contrastive Regularization uses triplets in a learned feature space where the denoised image is the anchor, the clean image is the positive sample, and the noisy image is the negative sample (Cui et al., 2023). The method trains a denoiser such as FastDVDnet on 5 noisy RAW images to output one denoised RAW image, then applies a contrastive term through Wnet, a task-specific feature embedding network based on Haar wavelet transform high-frequency subbands $$\min S_{\text{inner}}, \qquad \max S_{\text{inter}}.$$3, $$\min S_{\text{inner}}, \qquad \max S_{\text{inter}}.$$4, and $$\min S_{\text{inner}}, \qquad \max S_{\text{inter}}.$$5. The final objective is

$$\min S_{\text{inner}}, \qquad \max S_{\text{inter}}.$$6

and the strongest reported starlight result is 30.5 PSNR, 0.889 SSIM, and 0.060 LPIPS, which the paper describes as state-of-the-art performance on the starlight dataset.

In brain-to-image reconstruction, MindEye separates retrieval and reconstruction into parallel submodules: a contrastively trained fMRI-to-CLIP mapper for retrieval and a diffusion prior for image generation (Scotti et al., 2023). The model reports 93.6% image retrieval and 90.1% brain retrieval on the main averaged results, and the paper argues that separate heads are necessary because retrieval wants a discriminative embedding while reconstruction wants a generator-compatible latent. Contrast and Diffuse (CnD) likewise uses a two-phase design, first learning fMRI representations by self-supervised contrastive learning and then conditioning a latent diffusion model with a concept-aware module; it reports a 50-way accuracy improvement of about +7.7 over SS-AE and +3.7 over fMRI-ICGAN on GOD (Sun et al., 2023).

A more formal connection appears in Contrastive Diffusion Loss (CDL), which argues that a diffusion denoiser implicitly defines a classifier between noise levels and uses this observation to train the denoiser in out-of-distribution regions encountered during sampling (Wu et al., 2024). The key contrastive objective is

$$\min S_{\text{inner}}, \qquad \max S_{\text{inter}}.$$7

The reported improvements include lower parallel-sampling FID for CIFAR-10, DDPM from 9.43 to 9.06 and for FFHQ, VP from 6.26 to 5.89, supporting a principled contrastive-reconstruction view of diffusion training.

At the conditioning level, Diff-CA trains a prompt-free image-conditioned diffusion generator and then decomposes its conditioning tokens into common and salient parts using weak binary supervision (Soumm et al., 4 Jun 2026). Its additive factorization

$$\min S_{\text{inner}}, \qquad \max S_{\text{inter}}.$$8

is paired with pinning, cycle consistency, and adversarial purification, and the paper proves identifiability under mild assumptions. The method reports about 94.5% glasses swap accuracy while preserving identity, gender, smile, and pose, making it an explicit example of diffusion-based reconstruction with contrastive structure.

Finally, some papers use DCR for concepts that are diffusion-related but not “Diffusion Contrastive Reconstruction.” Default Completion Repulsion is a training-free guidance method for rare compositional generation that constructs a counterfactual attractor prompt and removes guidance aligned with the resulting drift direction (Kang et al., 7 May 2026). On DCR-Bench, the full method reports CLIPScore 0.3131, CLIP-attr 0.2558, BLIP 0.8075, CCS 4.13, and CVR 0.31. By contrast, in synthetic-data privacy, Distance to Closest Record is a nearest-neighbor similarity proxy rather than a reconstruction method, and the associated paper argues that it fails to identify actual privacy leakage as measured by membership inference attacks (Yao et al., 2 May 2025).

Across these uses, the common thread is not a single canonical algorithm but a recurring attempt to align contrastive structure with reconstruction or generation. In the narrow sense, Diffusion Contrastive Reconstruction refers to methods that use diffusion reconstruction as the substrate on which contrastive information is imposed; in the broader literature, closely related designs either separate contrastive and diffusion modules, reinterpret diffusion as contrastive discrimination across noise levels, or use the DCR acronym for entirely different objectives.