<pre id="j9nnl"></pre>
        <ruby id="j9nnl"></ruby>
          <pre id="j9nnl"></pre><ruby id="j9nnl"><mark id="j9nnl"></mark></ruby>
          <p id="j9nnl"><mark id="j9nnl"><progress id="j9nnl"></progress></mark></p>

              <p id="j9nnl"><cite id="j9nnl"></cite></p>

                  <ruby id="j9nnl"></ruby>

                  <del id="j9nnl"><dfn id="j9nnl"><th id="j9nnl"></th></dfn></del>

                        <ruby id="j9nnl"><mark id="j9nnl"><thead id="j9nnl"></thead></mark></ruby>
                        <p id="j9nnl"></p>

                        <p id="j9nnl"><mark id="j9nnl"></mark></p>
                          <p id="j9nnl"><del id="j9nnl"><thead id="j9nnl"></thead></del></p>

                          科學研究

                          Research

                          首頁 >  論文  > 詳情

                          CLIP-Adapter: Better Vision-Language Models with Feature Adapters

                          發表會議及期刊:arXiv

                          Peng Gao?1, Shijie Geng?2, Renrui Zhang?1, Teli Ma1, Rongyao Fang3

                          Yongfeng Zhang2, Hongsheng Li3, Yu Qiao

                          1Shanghai AI Laboratory     2Rutgers University

                           3The Chinese University of Hong Kong

                           {gaopeng,zhangrenrui,qiaoyu}@pjlab.org.cn 

                          sg1309@rutgers.edu,  hsli@ee.cuhk.edu.hk


                          Abstract:

                          Large-scale contrastive vision-language pretraining has shown significant progress in visual representation learning. Unlike traditional visual systems trained by a fixed set of discrete labels, a new paradigm was introduced in (Radford et al., 2021) to directly learn to align images with raw texts in an open-vocabulary setting. On downstream tasks, a carefully chosen text prompt is employed to make zero-shot predictions. To avoid non-trivial prompt engineering, context optimization (Zhou et al., 2021) has been proposed to learn continuous vectors as task-specific prompts with few-shot training examples. In this paper, we show that there is an alternative path to achieve better vision-language models other than prompt tuning. While prompt tuning is for the textual inputs, we propose CLIP-Adapter to conduct fine-tuning with feature adapters on either visual or language branch. Specifically, CLIP-Adapter adopts an additional bottleneck layer to learn new features and performs residual style feature blending with the original pre-trained features. As a consequence, CLIP-Adapter is able to outperform context optimization while maintains a simple design. Experiments and extensive ablation studies on various visual classification tasks demonstrate the effectiveness of our approach.

                          comm@pjlab.org.cn

                          上海市徐匯區云錦路701號西岸國際人工智能中心37-38層

                          滬ICP備2021009351號-1

                          <pre id="j9nnl"></pre>
                                <ruby id="j9nnl"></ruby>
                                  <pre id="j9nnl"></pre><ruby id="j9nnl"><mark id="j9nnl"></mark></ruby>
                                  <p id="j9nnl"><mark id="j9nnl"><progress id="j9nnl"></progress></mark></p>

                                      <p id="j9nnl"><cite id="j9nnl"></cite></p>

                                          <ruby id="j9nnl"></ruby>

                                          <del id="j9nnl"><dfn id="j9nnl"><th id="j9nnl"></th></dfn></del>

                                                <ruby id="j9nnl"><mark id="j9nnl"><thead id="j9nnl"></thead></mark></ruby>
                                                <p id="j9nnl"></p>

                                                <p id="j9nnl"><mark id="j9nnl"></mark></p>
                                                  <p id="j9nnl"><del id="j9nnl"><thead id="j9nnl"></thead></del></p>
                                                  韩国伦理电影