Generative object compositing methods have shown remarkable ability to seamlessly insert objects into scenes. However, when applied to real-world catalog image generation, these methods require tedious manual intervention: users must carefully adjust masks when product dimensions differ, and painstakingly restore occluded elements post-generation. We present CatalogStitch, a set of model-agnostic techniques that automate these corrections, enabling user-friendly content creation. Our dimension-aware mask computation algorithm automatically adapts the target region to accommodate products with different dimensions; users simply provide a product image and background, without manual mask adjustments. Our occlusion-aware hybrid restoration method guarantees pixel-perfect preservation of occluding elements, eliminating post-editing workflows. We additionally introduce CatalogStitch-Eval, a 58-example benchmark covering aspect-ratio mismatch and occlusion-heavy catalog scenarios, together with supplementary PDF and HTML viewers. We evaluate our techniques with three state-of-the-art compositing models (ObjectStitch, OmniPaint, and InsertAnything), demonstrating consistent improvements across diverse catalog scenarios. By reducing manual intervention and automating tedious corrections, our approach transforms generative compositing into a practical, human-friendly tool for production catalog workflows.
Figures & tables
Figure 1: Challenge 1: Product Dimension Mismatch. When replacing a product with different proportions, freeform and bounding box masks distort the product shape. Our dimension-aware mask preserves correct proportions.
Figure 2: Challenge 2: Occlusion Destruction. Foreground elements (table/decor items) occluding the target product are corrupted during compositing. Our hybrid restoration preserves them pixel-perfectly.
Figure 3: CatalogStitch overview. Two lightweight, model-agnostic modules wrap a baseline compositor. The target mask is adapted to the replacement product ratio; occluders are segmented, cached, and inpainted away before compositing, then restored from the original pixels at the final step.
Figure 4: Module-level flow charts. Left: dimension-aware mask computation preserves the original placement when the mismatch is small and otherwise expands the target region around the original centroid. Right: occlusion-aware restoration detects overlapping entities, caches their exact pixels, applies generative inpainting to remove occluders and expose a clean background, runs the compositor on the clean background, and finally restores the cached occluder pixels to preserve original geometry, texture, and scene-consistent lighting.
Figure 5: Dimension-aware mask computation across models. Left inputs: background and product. Mask overlay visualizes the dim-aware (blue) and bounding-box (pink) regions. Right: outputs under Freeform, BBox, and Dim-Aware masks for OmniPaint, InsertAnything, and ObjectStitch. Dim-aware masks preserve object proportions more reliably than Freeform and BBox masks across all models.
Figure 6: Occlusion-aware hybrid restoration across models. Left inputs: background and product. Mask overlay and occluder composite are visualizations of the occlusion regions. Right: Freeform, BBox, and Dim-Aware outcomes for InsertAnything and ObjectStitch, shown before and after occluder restoration. Our pipeline segments occluders, caches their pixels, inpaints them away to expose a clean background for compositing, and finally pastes the cached pixels back—preserving overlapping structures more faithfully than the corresponding before-restoration outputs.
Method
AR Error ↓
Occluder PSNR ↑
FID ↓
CLIP-score ↑
DINO-score ↑
OmniPaint
31.07
-
137.15
83.03
68.50
OmniPaint + Ours
4.57
-
135.79
86.11
72.99
ObjectStitch
30.97
11.60
101.55
90.27
85.76
ObjectStitch + Ours
5.05
26.84
91.52
90.62
88.09
InsertAnything
29.98
13.33
105.99
90.23
82.63
InsertAnything + Ours
3.92
27.54
77.72
92.68
88.30
Table 1: Quantitative comparison on CatalogStitch-Eval. AR Error, CLIP, and DINO scores are computed on the 35 dimension examples; Occluder PSNR on the 23 occlusion examples (– for OmniPaint). CLIP and DINO are cosine similarity ×100 ; FID is between input-product and generated-object crop distributions.
Despite strong single-turn performance, diffusion-based image compositing often struggles to preserve coherent spatial relations in pairwise or sequential edits, where subsequent insertions may overwrite previously generated content and disrupt physical consistency. We introduce PICS, a self-supervised composition-by-decomposition paradigm that composes objects in parallel while explicitly modeling the compositional interactions among (fully-/partially-)visible objects and background. At its core, an Interaction Transformer employs mask-guided Mixture-of-Experts to route background, exclusive, and overlap regions to dedicated experts, with an adaptive α-blending strategy that infers a compatibility-aware fusion of overlapping objects while preserving boundary fidelity. To further enhance robustness to geometric variations, we incorporate geometry-aware augmentations covering both out-of-plane and in-plane pose changes of objects. Our method delivers superior pairwise compositing quality and substantially improved stability, with extensive evaluations across virtual try-on, indoor, and street scene settings showing consistent gains over state-of-the-art baselines. Code and data are available at https://github.com/RyanHangZhou/PICS
Generative image composition aims to regenerate the given foreground object in the background image to produce a realistic composite image. Some high-authenticity methods can adjust foreground pose/view to be compatible with background, while some high-fidelity methods can preserve the foreground details accurately. However, existing methods can hardly achieve both goals at the same time. In this work, we propose a two-stage strategy to achieve both goals. In the first stage, we use high-authenticity method to generate reasonable foreground shape, serving as the condition of high-fidelity method in the second stage. The experiments on MureCOM dataset verify the effectiveness of our two-stage strategy. The code and model have been released at https://github.com/bcmi/OSInsert-Image-Composition.
Jingyuan Wang, Li Niu
MoE Key Lab of Artificial Intelligence, Shanghai Jiao Tong University
Automated product recognition is a cornerstone of modern retail intelligence; however, accurately matching real-world, in-store images against extensive corporate catalogs remains a major scalability bottleneck for large-scale applications. In this work, we address this challenge by reformulating the task as an embedding-based cross-domain retrieval problem rather than a standard closed-set classification task. Specifically, we define the objective as retrieving the most corresponding catalog reference image for a given real-world product query crop from an expansive inventory. To bridge the severe domain gap between pristine studio packshots and noisy in-store queries, we introduce a novel catalog-to-real multi-stage contrastive learning paradigm (Cat2Real). This framework fine-tunes a vision backbone by systematically exploiting both item-level and image-level similarities to drive targeted hard negative mining. Extensive empirical evaluations demonstrate that our paradigm scales seamlessly to unseen products and categories, yielding outstanding zero-shot generalization performance even in the complete absence of real-world training images for novel inventory.