Recommendation of Compatible Outfits Conditioned on Style

Banerjee, Debopriyo; Dhakad, Lucky; Maheshwari, Harsh; Chelliah, Muthusamy; Ganguly, Niloy; Bhattacharya, Arnab

doi:10.1007/978-3-030-99736-6_3

Cited by 2 publications

(4 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Context is important for fashion compatibility, even with the same partial outfit, different outfits can be recommended under different contexts. The context can be user perference [14]- [16], style category [2], body shape [17], scene image [1], or outfit text descripition. We propose to use an outfit text as context because it has advantages such as being easier to get than an image and more flexible to change than categorical style information.…”

Section: Related Work a Fashion Compatibility Recommendationmentioning

confidence: 99%

“…Without a text description, the number of possible outfits would be much larger. For the flexibility aspect, users can use natural language as the condition to compose an outfit instead of providing an image [1] or using limited style category [2], which is more convenient and intuitive. The benefit for society of building a better fashion recommendation system includes improving the shopping experience, improving business efficiency, empowering fashion economics, providing small brands more visibility, etc.…”

Section: Introductionmentioning

confidence: 99%

“…Recent fashion recommendation studies have explored complementary item retrieval (CIR) task [3] and styleconditioned outfit recommendation [2]. These tasks are similar to our task, but cannot achieve the goal of generating outfits from text with internal compatibility and condition compliance.…”

Section: Introductionmentioning

confidence: 99%

“…Moreover, most CIR approaches [3]- [6] do not consider the text condition, this may have little effect when there are many items are given as constraints, however, when there are few given items the generated outfit without text condition can be less accurate to users' need. Style-conditioned outfit recommendation [2] is conditioned on style categories with limited numbers, which is less flexible than outfit text description. Li [7] has the same goal as our work, but they implemented this by forcing each item representation, instead of an outfit representation, to be close to the outfit text description.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Text-Conditioned Outfit Recommendation With Hybrid Attention Layer

Wang,

Zhong

2024

IEEE Access

View full text Add to dashboard Cite

Text-conditioned outfit recommendation aims to recommend a whole fashion outfit that satisfies the compatibility between the recommended items and given items and adheres to the text condition like "Paradise Tropical Vacation" or "60s Style". Using text description as a condition can provide users with a flexible and accurate way to retrieve and recommend fashion items but this problem is underexplored by existing studies. A challenge of text-conditioned outfit recommendation is how to encode and fuse the outfit text description and fashion item images and text. To solve this, this paper proposes a framework for this task which features a hybrid attention layer that constructs the relationship between outfit text description and fashion items for condition compliance, and the relationship between fashion items for internal compatibility. To encode fashion item features, our method uses pre-trained FashionCLIP as an extractor which significantly reduces the trainable parameters compared to previous methods training CNN from scratch. The whole outfits are generated by iteratively adding compatible items based on a given partial outfit. Compared with state-of-the-art methods on polyvore disjoint and non-disjoint datasets, our approach can achieve 3% relative improvement in compatibility prediction AUC, achieve 5% relative improvement in fill-in-the-blank accuracy; achieve 19% relative improvement on complementary item retrieval recall at different ranks in average. Besides, We demonstrate that our approach can generate a whole outfit with inner compatibility and adhere to the text description.

show abstract