This is an repository that contains the resources for image-to-image translation (I2I) research.
-
Panoptic-aware Image-to-Image Translation [pdf]
-
UVCGAN: UNet Vision Transformer cycle-consistent GAN for unpaired image-to-image translation [pdf] [PyTorch]
-
Multi-domain image generation and translation with identifiability guarantees [OpenReview] [No code]
-
Diffusion-based Image Translation using disentangled style and content representation [OpenReview] [PyTorch]
-
Dual Diffusion Implicit Bridges for Image-to-Image Translation [OpenReview] [PyTorch]
- MIDMs: Matching Interleaved Diffusion Models for Exemplar-based Image Translation [pdf] [PyTorch]
- a diffusion-based matching-and-generation framework that interleaves cross-domain matching and diffusion in the latent space for I2I
-
EGSDE: Unpaired Image-to-Image Translation via Energy-Guided Stochastic Differential Equations [pdf] [code]
- an energy-guided stochastic differential equations that utilizes an energy function pretrained on source and target domains to guide the SDE inference for unpaired I2I
-
Unsupervised Image-to-Image Translation with Density Changing Regularization [pdf] [code]
- an unsupervised I2I model based on a density changing assumption that we should match image patches of high probability density for different domains.
-
Multi-Curve Translator for High-Resolution Photorealistic Image Translation [pdf] [code]
- a Multi-Curve Translator which predicts both the individual pixels and the neighor pixels for high-resolution I2I.
-
ManiFest: Manifold Deformation for Few-shot Image Translation [pdf] [code]
- a few-shot image translation model that learns a context-aware representation of a target domain using a style manifold between source and proxy anchor domains.
-
Vector Quantized Image-to-Image Translation [pdf] [code]
- A I2I framework based on vector quantized content representation
-
Unpaired Image Translation via Vector Symbolic Architectures [pdf] [code]
- a I2I framework based on Vector Symbolic Architectures which defines algebraic operations in a hypervector space.
-
VecGAN: Image-to-Image Translation with Interpretable Latent Directions [pdf] [NO CODE]
- a I2I framework with interpretable latent directions using latent space factorization and controllable strength of change.
-
Bi-level Feature Alignment for Versatile Image Translation and Manipulation [pdf] [code]
- a I2I framework using a bi-level feature alignment strategy that adopts a top-k operation to rank block-wise features and dense attention between block features to reduce memory cost.
-
Exploring Patch-Wise Semantic Relation for Contrastive Learning in Image-to-Image Translation Tasks [pdf] [code]
- a I2I framework based on semantic relation consistency and regularization along with the decoupled contrastive learning
-
Alleviating Semantics Distortion in Unsupervised Low-Level Image-to-Image Translation via Structure Consistency Constraint [pdf] [code]
- a Structure Consistency Constraint that reduces the randomness of color transformation in I2I.
-
A Style-Aware Discriminator for Controllable Image Translation [pdf] [code]
- a style-aware discriminator that acts as both the critic and the style encoder to provide conditions for the generator in I2I.
-
Wavelet Knowledge Distillation: Towards Efficient Image-to-Image Translation [pdf] [NO CODE]
- a I2I method based on high frequency bands distillation from discrete wavelet transformation.
-
InstaFormer: Instance-Aware Image-to-Image Translation With Transformer [pdf] [code]
- a transformer-based architecture with with adaptive instance normalization for instance-aware I2I.
-
Maximum Spatial Perturbation Consistency for Unpaired Image-to-Image Translation [pdf] [code]
- a universal regularization technique for I2I called maximum spatial perturbation consistency which enforces the spatial perturbation function and translation operator to be commutative.
-
FlexIT: Towards Flexible Semantic Image Translation [pdf] [code]
- a semantic image translation method based on autoencoder latent space and multi-modal embedding space
-
Self-Supervised Dense Consistency Regularization for Image-to-Image Translation [pdf] [NO CODE]
- an auxiliary self-supervision loss with dense consistency regularization for I2I.
-
Unsupervised Image-to-Image Translation With Generative Prior [pdf] [code]
- a I2I framework that uses the generative prior from GANs to learn rich content correspondences across various domains
-
QS-Attn: Query-Selected Attention for Contrastive Learning in I2I Translation [pdf] [code]
- a I2I framework based on a query-selected attention module, which compares feature distances in the source domain and select queries acc. to the measurement of signficance.
-
OA-FSUI2IT: A Novel Few-Shot Cross Domain Object Detection Framework with Object-Aware Few-Shot Unsupervised Image-to-Image Translation [pdf] [NO CODE]
- an Object-Aware Few-Shot Image Translation framework for few-shot cross domain object detection
-
Style-Guided and Disentangled Representation for Robust Image-to-Image Translation [pdf] [NO CODE]
- a I2I framework with a style-guided disriminator using flexible decision boundary and independent domain attributes
-
Image-to-image translation: Methods and applications. 2021 [pdf]
- lightweight network design for better efficiency
- generalize to cross-modality tasks (e.g., NLP, speech)
-
Deep Generative Adversarial Networks for Image-to-Image Translation: A Review paper. 2020 [pdf]
- Solve mode collapse
- More realistic evaluation metrics
- More image diversity
- Deep Reinforcement Learning
- 3D image-to-image translation
- 3D datasets
- Cybersecurity applications
-
An Overview of Image-to-Image Translation Using Generative Adversarial Networks [pdf]
- Combine GAN with other methods (e.g., VAE) to stabilize training
- GAN compression for lightweight design
- Transfer other methods (e.g., SR, Attention, OT) to I2IT
- Remove unnecessary components
- Extend to video
-
Unsupervised Image-to-Image Translation: A Review [pdf]
-
Applications of I2I to rainy days
-
Domain Bridge for Unpaired Image-to-Image Translation and Unsupervised Domain Adaptation [pdf]
-
Closing the Loop: Joint Rain Generation and Removal via Disentangled Image Translation [pdf]
-
From Rain Generation to Rain Removal [pdf]
-
DerainCycleGAN: Rain Attentive CycleGAN for Single Image Deraining and Rainmaking [pdf]
-
Close the Loop: A Unified Bottom-up and Top-down Paradigm for Joint Image Deraining and Segmentation [pdf]
-
-
Amazon Mechanical Turk (AMT)
-
Peak Signal-to-Noise Ratio (PSNR) ↑
-
Structural Similarity Index Measure (SSIM) ↑
-
Inception Score (IS) ↑
-
Fréchet Inception Distance (FID) ↓
-
Kernel Inception Distance (KID) ↓
-
Perceptual Distance (PD) ↓
-
Learned Perceptual Image Patch Similarity (LPIPS) ↓
-
FCN ↑
-
Density and Coverage (DC) ↑