arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2112.09445cs.CV
收藏

采用最优传输蒸馏的数据高效语言监督零样本识别

Data Efficient Language-supervised Zero-shot Recognition with Optimal Transport Distillation

Meta Reality Labs(Meta现实实验室);UC Berkeley(加州大学伯克利分校)

Bichen Wu, Ruizhe Cheng, Peizhao Zhang, Tianren Gao, Peter Vajda, Joseph E. Gonzalez

AI总结:

针对 CLIP 依赖超 400M 噪声图文对的问题,本文提出 OTTER,利用在线熵正则化最优传输生成软匹配标签进行对比蒸馏,仅用 3M 图文对就在多个大规模零样本基准上显著优于基线。

摘要:

传统计算机视觉模型被训练用于预测一组固定的预定义类别。近来,自然语言已被证明是一种更广泛、更丰富的监督来源,相比受监督的“黄金”标签,它能为视觉概念提供更细致的描述。以往工作(如 CLIP)使用 InfoNCE 损失训练模型,以预测图像与文本说明之间的配对。然而,CLIP 对数据需求很大,需要超过 400M 个图文对进行训练。这种低效部分可归因于图文对存在噪声。为解决该问题,我们提出 OTTER(Optimal TransporT distillation for Efficient zero-shot Recognition,用于高效零样本识别的最优传输蒸馏),它使用在线熵正则化最优传输来寻找软图文匹配,作为对比学习的标签。基于预训练的图像和文本编码器,使用 OTTER 训练的模型仅需 3M 个图文对即可取得强劲性能。与 InfoNCE 损失、标签平滑和知识蒸馏相比,OTTER 在 Google Open Images(19,958 个类别)以及来自 Tencent ML-Images 的多标签 ImageNet 10K(10032 个类别)的零样本评估中始终优于这些基线。在 7 种不同数据集/架构设置 × 6 项指标、共计超过 42 项评估中,OTTER 在其中 34 项优于(32 项)或持平(2 项)所有基线。

英文摘要:

Traditional computer vision models are trained to predict a fixed set of predefined categories. Recently, natural language has been shown to be a broader and richer source of supervision that provides finer descriptions to visual concepts than supervised "gold" labels. Previous works, such as CLIP, use InfoNCE loss to train a model to predict the pairing between images and text captions. CLIP, however, is data hungry and requires more than 400M image-text pairs for training. The inefficiency can be partially attributed to the fact that the image-text pairs are noisy. To address this, we propose OTTER (Optimal TransporT distillation for Efficient zero-shot Recognition), which uses online entropic optimal transport to find a soft image-text match as labels for contrastive learning. Based on pretrained image and text encoders, models trained with OTTER achieve strong performance with only 3M image text pairs. Compared with InfoNCE loss, label smoothing, and knowledge distillation, OTTER consistently outperforms these baselines in zero shot evaluation on Google Open Images (19,958 classes) and multi-labeled ImageNet 10K (10032 classes) from Tencent ML-Images. Over 42 evaluations on 7 different dataset/architecture settings x 6 metrics, OTTER outperforms (32) or ties (2) all baselines in 34 of them.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑