CLIP-Decoder : ZeroShot Multilabel Classification using Multimodal CLIP Aligned Representation

TL;DR

The paper introduces CLIP-Decoder, a multimodal extension of ML-Decoder that learns CLIP-aligned representations for zero-shot multi-label classification. By converting class labels into CLIP-prompts and fusing image features from a TResNet with text features from a CLIP encoder through a shared projection, it optimizes a joint loss $L = \alpha L_{clip} + \beta L_c$ to align modalities. The approach demonstrates state-of-the-art results on ZSL and GZSL benchmarks (notably on NUS-WIDE), achieving significant absolute gains in mAP over prior methods and showing robustness across prompt designs and embeddings. This method offers practical impact for recognizing unseen labels in complex, real-world multi-label tasks and points to extensions in domains like zero-shot action recognition and multiomics analyses.

Abstract

Multi-label classification is an essential task utilized in a wide variety of real-world applications. Multi-label zero-shot learning is a method for classifying images into multiple unseen categories for which no training data is available, while in general zero-shot situations, the test set may include observed classes. The CLIP-Decoder is a novel method based on the state-of-the-art ML-Decoder attention-based head. We introduce multi-modal representation learning in CLIP-Decoder, utilizing the text encoder to extract text features and the image encoder for image feature extraction. Furthermore, we minimize semantic mismatch by aligning image and word embeddings in the same dimension and comparing their respective representations using a combined loss, which comprises classification loss and CLIP loss. This strategy outperforms other methods and we achieve cutting-edge results on zero-shot multilabel classification tasks using CLIP-Decoder. Our method achieves an absolute increase of 3.9% in performance compared to existing methods for zero-shot learning multi-label classification tasks. Additionally, in the generalized zero-shot learning multi-label classification task, our method shows an impressive increase of almost 2.3%.

Figures (3)

• Figure 1: Comparison of proposed method with the state-of-the-art method in multi-label zero shot learning, ML-Decoder RidnikML-Decoder:Head.
• Figure 2: Architecture of the proposed method, utilizing a TResNet image encoder for image embeddings and a CLIP text encoder for feature representations. The method integrates classification and CLIP loss for enhanced alignment
• Figure 3: Multi-label classification in GZSL.