{"id":479505,"date":"2023-08-09T10:41:18","date_gmt":"2023-08-09T10:41:18","guid":{"rendered":""},"modified":"2023-09-05T11:18:58","modified_gmt":"2023-09-05T11:18:58","slug":"vector-quantized-generative-adversarial-network-vqgan","status":"publish","type":"wiki","link":"https:\/\/oneproxy.pro\/cn\/wiki\/vector-quantized-generative-adversarial-network-vqgan\/","title":{"rendered":"\u77e2\u91cf\u91cf\u5316\u751f\u6210\u5bf9\u6297\u7f51\u7edc (VQGAN)"},"content":{"rendered":"

\u77e2\u91cf\u91cf\u5316\u751f\u6210\u5bf9\u6297\u7f51\u7edc (VQGAN) \u662f\u4e00\u79cd\u521b\u65b0\u4e14\u5f3a\u5927\u7684\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\uff0c\u5b83\u7ed3\u5408\u4e86\u4e24\u79cd\u6d41\u884c\u673a\u5668\u5b66\u4e60\u6280\u672f\u7684\u5143\u7d20\uff1a\u751f\u6210\u5bf9\u6297\u7f51\u7edc (GAN) \u548c\u77e2\u91cf\u91cf\u5316 (VQ)\u3002VQGAN \u56e0\u5176\u80fd\u591f\u751f\u6210\u9ad8\u8d28\u91cf\u4e14\u8fde\u8d2f\u7684\u56fe\u50cf\u800c\u5f15\u8d77\u4e86\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u754c\u7684\u6781\u5927\u5173\u6ce8\uff0c\u4f7f\u5176\u6210\u4e3a\u5404\u79cd\u5e94\u7528\u7684\u6709\u524d\u9014\u7684\u5de5\u5177\uff0c\u5305\u62ec\u56fe\u50cf\u5408\u6210\u3001\u98ce\u683c\u8f6c\u6362\u548c\u521b\u610f\u5185\u5bb9\u751f\u6210\u3002<\/p>\n

\u77e2\u91cf\u91cf\u5316\u751f\u6210\u5bf9\u6297\u7f51\u7edc\uff08VQGAN\uff09\u7684\u8d77\u6e90\u5386\u53f2\u4ee5\u53ca\u9996\u6b21\u63d0\u53ca\u5b83\u3002<\/h2>\n

GAN \u7684\u6982\u5ff5\u6700\u65e9\u7531 Ian Goodfellow \u53ca\u5176\u540c\u4e8b\u4e8e 2014 \u5e74\u63d0\u51fa\u3002GAN \u662f\u4e00\u79cd\u751f\u6210\u6a21\u578b\uff0c\u7531\u4e24\u4e2a\u795e\u7ecf\u7f51\u7edc\uff08\u751f\u6210\u5668\u548c\u9274\u522b\u5668\uff09\u7ec4\u6210\uff0c\u5b83\u4eec\u901a\u8fc7\u6781\u5c0f\u6781\u5927\u535a\u5f08\u6765\u751f\u6210\u903c\u771f\u7684\u5408\u6210\u6570\u636e\u3002\u867d\u7136 GAN \u5728\u751f\u6210\u56fe\u50cf\u65b9\u9762\u8868\u73b0\u51fa\u8272\uff0c\u4f46\u5b83\u4eec\u53ef\u80fd\u4f1a\u51fa\u73b0\u6a21\u5f0f\u5d29\u6e83\u548c\u5bf9\u751f\u6210\u8f93\u51fa\u7f3a\u4e4f\u63a7\u5236\u7b49\u95ee\u9898\u3002<\/p>\n

2020 \u5e74\uff0cDeepMind \u7684\u7814\u7a76\u4eba\u5458\u63a8\u51fa\u4e86\u77e2\u91cf\u91cf\u5316\u53d8\u5206\u81ea\u7f16\u7801\u5668 (VQ-VAE) \u6a21\u578b\u3002VQ-VAE \u662f\u53d8\u5206\u81ea\u7f16\u7801\u5668 (VAE) \u6a21\u578b\u7684\u4e00\u79cd\u53d8\u4f53\uff0c\u5b83\u7ed3\u5408\u4e86\u77e2\u91cf\u91cf\u5316\u6765\u751f\u6210\u8f93\u5165\u6570\u636e\u7684\u79bb\u6563\u4e14\u7d27\u51d1\u7684\u8868\u793a\u3002\u8fd9\u662f VQGAN \u5f00\u53d1\u7684\u5173\u952e\u4e00\u6b65\u3002<\/p>\n

\u968f\u540e\uff0c\u540c\u5e74\uff0c\u7531 Ali Razavi \u9886\u5bfc\u7684\u4e00\u7ec4\u7814\u7a76\u4eba\u5458\u63a8\u51fa\u4e86 VQGAN\u3002\u8be5\u6a21\u578b\u7ed3\u5408\u4e86 GAN \u7684\u5f3a\u5927\u529f\u80fd\u548c VQ-VAE \u7684\u77e2\u91cf\u91cf\u5316\u6280\u672f\uff0c\u53ef\u4ee5\u751f\u6210\u8d28\u91cf\u3001\u7a33\u5b9a\u6027\u548c\u63a7\u5236\u6027\u5747\u6709\u6240\u63d0\u9ad8\u7684\u56fe\u50cf\u3002VQGAN \u6210\u4e3a\u751f\u6210\u6a21\u578b\u9886\u57df\u7684\u4e00\u9879\u7a81\u7834\u6027\u8fdb\u6b65\u3002<\/p>\n

\u6709\u5173\u77e2\u91cf\u91cf\u5316\u751f\u6210\u5bf9\u6297\u7f51\u7edc (VQGAN) \u7684\u8be6\u7ec6\u4fe1\u606f\u3002\u6269\u5c55\u77e2\u91cf\u91cf\u5316\u751f\u6210\u5bf9\u6297\u7f51\u7edc (VQGAN) \u4e3b\u9898\u3002<\/h2>\n

\u77e2\u91cf\u91cf\u5316\u751f\u6210\u5bf9\u6297\u7f51\u7edc (VQGAN) \u7684\u5de5\u4f5c\u539f\u7406<\/h3>\n

VQGAN \u548c\u4f20\u7edf GAN \u4e00\u6837\uff0c\u7531\u751f\u6210\u5668\u548c\u9274\u522b\u5668\u7ec4\u6210\u3002\u751f\u6210\u5668\u4ee5\u968f\u673a\u566a\u58f0\u4f5c\u4e3a\u8f93\u5165\uff0c\u5e76\u5c1d\u8bd5\u751f\u6210\u903c\u771f\u7684\u56fe\u50cf\uff0c\u800c\u9274\u522b\u5668\u5219\u65e8\u5728\u533a\u5206\u771f\u5b9e\u56fe\u50cf\u548c\u751f\u6210\u7684\u56fe\u50cf\u3002<\/p>\n

VQGAN \u7684\u5173\u952e\u521b\u65b0\u5728\u4e8e\u5176\u7f16\u7801\u5668\u67b6\u6784\u3002\u7f16\u7801\u5668\u4e0d\u4f7f\u7528\u8fde\u7eed\u8868\u793a\uff0c\u800c\u662f\u5c06\u8f93\u5165\u56fe\u50cf\u6620\u5c04\u5230\u79bb\u6563\u6f5c\u7801\uff0c\u8868\u793a\u56fe\u50cf\u7684\u4e0d\u540c\u5143\u7d20\u3002\u7136\u540e\uff0c\u8fd9\u4e9b\u79bb\u6563\u4ee3\u7801\u901a\u8fc7\u5305\u542b\u4e00\u7ec4\u9884\u5b9a\u4e49\u5d4c\u5165\u6216\u5411\u91cf\u7684\u7801\u672c\u3002\u7801\u672c\u4e2d\u6700\u63a5\u8fd1\u7684\u5d4c\u5165\u5c06\u66ff\u6362\u539f\u59cb\u4ee3\u7801\uff0c\u4ece\u800c\u4ea7\u751f\u91cf\u5316\u8868\u793a\u3002\u6b64\u8fc7\u7a0b\u79f0\u4e3a\u77e2\u91cf\u91cf\u5316\u3002<\/p>\n

\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u7f16\u7801\u5668\u3001\u751f\u6210\u5668\u548c\u9274\u522b\u5668\u534f\u4f5c\u4ee5\u6700\u5c0f\u5316\u91cd\u5efa\u635f\u5931\u548c\u5bf9\u6297\u635f\u5931\uff0c\u4ece\u800c\u786e\u4fdd\u751f\u6210\u4e0e\u8bad\u7ec3\u6570\u636e\u76f8\u4f3c\u7684\u9ad8\u8d28\u91cf\u56fe\u50cf\u3002VQGAN \u4f7f\u7528\u79bb\u6563\u6f5c\u7801\u589e\u5f3a\u4e86\u5176\u6355\u83b7\u6709\u610f\u4e49\u7ed3\u6784\u7684\u80fd\u529b\uff0c\u5e76\u5b9e\u73b0\u4e86\u66f4\u53ef\u63a7\u7684\u56fe\u50cf\u751f\u6210\u3002<\/p>\n

\u77e2\u91cf\u91cf\u5316\u751f\u6210\u5bf9\u6297\u7f51\u7edc (VQGAN) \u7684\u4e3b\u8981\u7279\u70b9<\/h3>\n
    \n
  1. \n

    \u79bb\u6563\u6f5c\u7801<\/strong>\uff1aVQGAN \u91c7\u7528\u79bb\u6563\u6f5c\u5728\u4ee3\u7801\uff0c\u4f7f\u5176\u80fd\u591f\u4ea7\u751f\u591a\u6837\u5316\u4e14\u53ef\u63a7\u7684\u56fe\u50cf\u8f93\u51fa\u3002<\/p>\n<\/li>\n

  2. \n

    \u5c42\u6b21\u7ed3\u6784<\/strong>\uff1a\u6a21\u578b\u7684\u7801\u672c\u5f15\u5165\u4e86\u5206\u5c42\u7ed3\u6784\uff0c\u589e\u5f3a\u4e86\u8868\u793a\u5b66\u4e60\u8fc7\u7a0b\u3002<\/p>\n<\/li>\n

  3. \n

    \u7a33\u5b9a<\/strong>\uff1aVQGAN \u89e3\u51b3\u4e86\u4f20\u7edf GAN \u4e2d\u89c2\u5bdf\u5230\u7684\u4e00\u4e9b\u4e0d\u7a33\u5b9a\u95ee\u9898\uff0c\u4ece\u800c\u5b9e\u73b0\u66f4\u6d41\u7545\u3001\u66f4\u4e00\u81f4\u7684\u8bad\u7ec3\u3002<\/p>\n<\/li>\n

  4. \n

    \u9ad8\u8d28\u91cf\u56fe\u50cf\u751f\u6210<\/strong>\uff1aVQGAN \u53ef\u4ee5\u751f\u6210\u9ad8\u5206\u8fa8\u7387\u3001\u5177\u6709\u4ee4\u4eba\u5370\u8c61\u6df1\u523b\u7684\u7ec6\u8282\u548c\u8fde\u8d2f\u6027\u7684\u89c6\u89c9\u5438\u5f15\u529b\u7684\u56fe\u50cf\u3002<\/p>\n<\/li>\n<\/ol>\n

    \u77e2\u91cf\u91cf\u5316\u751f\u6210\u5bf9\u6297\u7f51\u7edc (VQGAN) \u7684\u7c7b\u578b<\/h2>\n

    VQGAN \u81ea\u8bde\u751f\u4ee5\u6765\u4e00\u76f4\u5728\u4e0d\u65ad\u53d1\u5c55\uff0c\u5e76\u4e14\u5df2\u7ecf\u63d0\u51fa\u4e86\u591a\u79cd\u53d8\u4f53\u548c\u6539\u8fdb\u3002\u4e00\u4e9b\u503c\u5f97\u6ce8\u610f\u7684 VQGAN \u7c7b\u578b\u5305\u62ec\uff1a<\/p>\n\n\n\n\n\n\n\n
    \u7c7b\u578b<\/th>\n\u63cf\u8ff0<\/th>\n<\/tr>\n<\/thead>\n
    \u7ef4\u57fa\u767e\u79d1<\/td>\n\u5bf9 VQ-VAE \u8fdb\u884c\u6269\u5c55\uff0c\u6539\u8fdb\u4e86\u77e2\u91cf\u91cf\u5316\u3002<\/td>\n<\/tr>\n
    VQGAN+CLIP<\/td>\n\u5c06 VQGAN \u4e0e CLIP \u6a21\u578b\u76f8\u7ed3\u5408\uff0c\u5b9e\u73b0\u66f4\u597d\u7684\u56fe\u50cf\u63a7\u5236\u3002<\/td>\n<\/tr>\n
    \u6269\u6563\u6a21\u578b<\/td>\n\u96c6\u6210\u6269\u6563\u6a21\u578b\u4ee5\u5b9e\u73b0\u9ad8\u8d28\u91cf\u56fe\u50cf\u5408\u6210\u3002<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u77e2\u91cf\u91cf\u5316\u751f\u6210\u5bf9\u6297\u7f51\u7edc\uff08VQGAN\uff09\u7684\u4f7f\u7528\u65b9\u6cd5\u3001\u95ee\u9898\u53ca\u5176\u4f7f\u7528\u76f8\u5173\u7684\u89e3\u51b3\u65b9\u6848\u3002<\/h2>\n

    \u77e2\u91cf\u91cf\u5316\u751f\u6210\u5bf9\u6297\u7f51\u7edc (VQGAN) \u7684\u7528\u9014<\/h3>\n
      \n
    1. \n

      \u56fe\u50cf\u5408\u6210<\/strong>\uff1aVQGAN \u53ef\u4ee5\u751f\u6210\u903c\u771f\u4e14\u591a\u6837\u5316\u7684\u56fe\u50cf\uff0c\u53ef\u7528\u4e8e\u521b\u610f\u5185\u5bb9\u751f\u6210\u3001\u827a\u672f\u548c\u8bbe\u8ba1\u3002<\/p>\n<\/li>\n

    2. \n

      \u98ce\u683c\u8f6c\u79fb<\/strong>\uff1a\u901a\u8fc7\u64cd\u7eb5\u6f5c\u5728\u4ee3\u7801\uff0cVQGAN \u53ef\u4ee5\u6267\u884c\u98ce\u683c\u8f6c\u6362\uff0c\u6539\u53d8\u56fe\u50cf\u7684\u5916\u89c2\uff0c\u540c\u65f6\u4fdd\u7559\u5176\u7ed3\u6784\u3002<\/p>\n<\/li>\n

    3. \n

      \u6570\u636e\u589e\u5f3a<\/strong>\uff1aVQGAN \u53ef\u7528\u4e8e\u589e\u5f3a\u5176\u4ed6\u8ba1\u7b97\u673a\u89c6\u89c9\u4efb\u52a1\u7684\u8bad\u7ec3\u6570\u636e\uff0c\u4ece\u800c\u63d0\u9ad8\u673a\u5668\u5b66\u4e60\u6a21\u578b\u7684\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n<\/li>\n<\/ol>\n

      \u95ee\u9898\u4e0e\u89e3\u51b3\u65b9\u6848<\/h3>\n
        \n
      1. \n

        \u8bad\u7ec3\u4e0d\u7a33\u5b9a<\/strong>\uff1a\u4e0e\u8bb8\u591a\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u4e00\u6837\uff0cVQGAN \u53ef\u80fd\u5b58\u5728\u8bad\u7ec3\u4e0d\u7a33\u5b9a\u7684\u95ee\u9898\uff0c\u5bfc\u81f4\u6a21\u5f0f\u5d29\u6e83\u6216\u6536\u655b\u6027\u8f83\u5dee\u3002\u7814\u7a76\u4eba\u5458\u901a\u8fc7\u8c03\u6574\u8d85\u53c2\u6570\u3001\u4f7f\u7528\u6b63\u5219\u5316\u6280\u672f\u548c\u5f15\u5165\u67b6\u6784\u6539\u8fdb\u6765\u89e3\u51b3\u6b64\u95ee\u9898\u3002<\/p>\n<\/li>\n

      2. \n

        \u7801\u672c\u5927\u5c0f<\/strong>\uff1a\u7801\u672c\u7684\u5927\u5c0f\u4f1a\u663e\u8457\u5f71\u54cd\u6a21\u578b\u7684\u5185\u5b58\u9700\u6c42\u548c\u8bad\u7ec3\u65f6\u95f4\u3002\u7814\u7a76\u4eba\u5458\u5df2\u7ecf\u63a2\u7d22\u4e86\u5728\u4e0d\u727a\u7272\u56fe\u50cf\u8d28\u91cf\u7684\u60c5\u51b5\u4e0b\u4f18\u5316\u7801\u672c\u5927\u5c0f\u7684\u65b9\u6cd5\u3002<\/p>\n<\/li>\n

      3. \n

        \u53ef\u63a7\u6027<\/strong>\uff1a\u867d\u7136 VQGAN \u5141\u8bb8\u5bf9\u56fe\u50cf\u751f\u6210\u8fdb\u884c\u4e00\u5b9a\u7a0b\u5ea6\u7684\u63a7\u5236\uff0c\u4f46\u5b9e\u73b0\u7cbe\u786e\u63a7\u5236\u4ecd\u7136\u5177\u6709\u6311\u6218\u6027\u3002\u7814\u7a76\u4eba\u5458\u6b63\u5728\u79ef\u6781\u7814\u7a76\u63d0\u9ad8\u6a21\u578b\u53ef\u63a7\u6027\u7684\u65b9\u6cd5\u3002<\/p>\n<\/li>\n<\/ol>\n

        \u4ee5\u8868\u683c\u548c\u5217\u8868\u7684\u5f62\u5f0f\u5217\u51fa\u4e3b\u8981\u7279\u5f81\u4ee5\u53ca\u4e0e\u7c7b\u4f3c\u672f\u8bed\u7684\u5176\u4ed6\u6bd4\u8f83\u3002<\/h2>\n

        \u4e0e\u4f20\u7edf GAN \u548c VAE \u7684\u6bd4\u8f83<\/h3>\n\n\n\n\n\n\n\n\n
        \u7279\u5f81<\/th>\n\u5411\u91cf\u751f\u6210\u5bf9\u6297\u7f51\u7edc<\/th>\n\u4f20\u7edf GAN<\/th>\n\u8840\u7ba1\u5185\u76ae\u7ec6\u80de<\/th>\n<\/tr>\n<\/thead>\n
        \u6f5c\u5728\u7a7a\u95f4\u8868\u793a<\/td>\n\u79bb\u6563\u4ee3\u7801<\/td>\n\u8fde\u7eed\u503c<\/td>\n\u8fde\u7eed\u503c<\/td>\n<\/tr>\n
        \u753b\u9762\u8d28\u91cf<\/td>\n\u9ad8\u8d28\u91cf<\/td>\n\u54c1\u8d28\u53c2\u5dee\u4e0d\u9f50<\/td>\n\u4e2d\u7b49\u8d28\u91cf<\/td>\n<\/tr>\n
        \u6a21\u5f0f\u5d29\u6e83<\/td>\n\u51cf\u5c11<\/td>\n\u5bb9\u6613\u5d29\u6e83<\/td>\n\u4e0d\u9002\u7528<\/td>\n<\/tr>\n
        \u53ef\u63a7\u6027<\/td>\n\u6539\u8fdb\u63a7\u5236<\/td>\n\u6709\u9650\u63a7\u5236<\/td>\n\u826f\u597d\u7684\u63a7\u5236<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

        \u4e0e\u5176\u4ed6\u751f\u6210\u6a21\u578b\u7684\u6bd4\u8f83<\/h3>\n\n\n\n\n\n\n\n
        \u6a21\u578b<\/th>\n\u7279\u5f81<\/th>\n\u5e94\u7528\u9886\u57df<\/th>\n<\/tr>\n<\/thead>\n
        \u5411\u91cf\u91cf\u5316<\/td>\n\u5728\u53d8\u5206\u81ea\u52a8\u7f16\u7801\u5668\u6846\u67b6\u4e2d\u4f7f\u7528\u77e2\u91cf\u91cf\u5316\u3002<\/td>\n\u56fe\u50cf\u538b\u7f29\u3001\u6570\u636e\u8868\u793a\u3002<\/td>\n<\/tr>\n
        \u5939\u5b50<\/td>\n\u89c6\u89c9\u548c\u8bed\u8a00\u9884\u8bad\u7ec3\u6a21\u578b\u3002<\/td>\n\u56fe\u50cf\u5b57\u5e55\u3001\u6587\u672c\u5230\u56fe\u50cf\u751f\u6210\u3002<\/td>\n<\/tr>\n
        \u6269\u6563\u6a21\u578b<\/td>\n\u56fe\u50cf\u5408\u6210\u7684\u6982\u7387\u6a21\u578b\u3002<\/td>\n\u9ad8\u8d28\u91cf\u56fe\u50cf\u751f\u6210\u3002<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

        \u4e0e\u77e2\u91cf\u91cf\u5316\u751f\u6210\u5bf9\u6297\u7f51\u7edc (VQGAN) \u76f8\u5173\u7684\u672a\u6765\u89c2\u70b9\u548c\u6280\u672f\u3002<\/h2>\n

        VQGAN \u5df2\u7ecf\u5728\u5404\u79cd\u521b\u610f\u5e94\u7528\u4e2d\u5c55\u73b0\u51fa\u5de8\u5927\u7684\u6f5c\u529b\uff0c\u5176\u672a\u6765\u524d\u666f\u5149\u660e\u3002\u4e0e VQGAN \u76f8\u5173\u7684\u4e00\u4e9b\u6f5c\u5728\u672a\u6765\u53d1\u5c55\u548c\u6280\u672f\u5305\u62ec\uff1a<\/p>\n

          \n
        1. \n

          \u63d0\u9ad8\u53ef\u63a7\u6027<\/strong>\uff1a\u7814\u7a76\u7684\u8fdb\u6b65\u53ef\u80fd\u4f1a\u5bf9\u751f\u6210\u7684\u56fe\u50cf\u8fdb\u884c\u66f4\u7cbe\u786e\u3001\u66f4\u76f4\u89c2\u7684\u63a7\u5236\uff0c\u4e3a\u827a\u672f\u8868\u8fbe\u5f00\u8f9f\u65b0\u7684\u53ef\u80fd\u6027\u3002<\/p>\n<\/li>\n

        2. \n

          \u591a\u6a21\u6001\u751f\u6210<\/strong>\uff1a\u7814\u7a76\u4eba\u5458\u6b63\u5728\u63a2\u7d22\u4f7f VQGAN \u80fd\u591f\u751f\u6210\u591a\u79cd\u98ce\u683c\u6216\u6a21\u5f0f\u7684\u56fe\u50cf\u7684\u65b9\u6cd5\uff0c\u4ece\u800c\u5b9e\u73b0\u66f4\u52a0\u591a\u6837\u5316\u548c\u5bcc\u6709\u521b\u610f\u7684\u8f93\u51fa\u3002<\/p>\n<\/li>\n

        3. \n

          \u5b9e\u65f6\u751f\u6210<\/strong>\uff1a\u968f\u7740\u786c\u4ef6\u548c\u4f18\u5316\u6280\u672f\u7684\u8fdb\u6b65\uff0c\u4f7f\u7528 VQGAN \u8fdb\u884c\u5b9e\u65f6\u56fe\u50cf\u751f\u6210\u53ef\u80fd\u4f1a\u53d8\u5f97\u66f4\u52a0\u53ef\u884c\uff0c\u4ece\u800c\u5b9e\u73b0\u4ea4\u4e92\u5f0f\u5e94\u7528\u3002<\/p>\n<\/li>\n<\/ol>\n

          \u5982\u4f55\u4f7f\u7528\u4ee3\u7406\u670d\u52a1\u5668\u6216\u5c06\u5176\u4e0e\u77e2\u91cf\u91cf\u5316\u751f\u6210\u5bf9\u6297\u7f51\u7edc (VQGAN) \u5173\u8054\u3002<\/h2>\n

          \u4ee3\u7406\u670d\u52a1\u5668\u5728\u652f\u6301 VQGAN \u7684\u4f7f\u7528\u65b9\u9762\u53ef\u4ee5\u53d1\u6325\u81f3\u5173\u91cd\u8981\u7684\u4f5c\u7528\uff0c\u5c24\u5176\u662f\u5728\u6d89\u53ca\u5927\u89c4\u6a21\u6570\u636e\u5904\u7406\u548c\u56fe\u50cf\u751f\u6210\u7684\u573a\u666f\u4e2d\u3002\u4ee5\u4e0b\u662f\u4ee3\u7406\u670d\u52a1\u5668\u53ef\u7528\u4e8e\u6216\u4e0e VQGAN \u5173\u8054\u7684\u4e00\u4e9b\u65b9\u6cd5\uff1a<\/p>\n

            \n
          1. \n

            \u6570\u636e\u6536\u96c6\u548c\u9884\u5904\u7406<\/strong>\uff1a\u4ee3\u7406\u670d\u52a1\u5668\u53ef\u4ee5\u5e2e\u52a9\u6536\u96c6\u548c\u9884\u5904\u7406\u6765\u81ea\u5404\u79cd\u6765\u6e90\u7684\u56fe\u50cf\u6570\u636e\uff0c\u786e\u4fdd\u7528\u4e8e\u8bad\u7ec3 VQGAN \u7684\u6570\u636e\u96c6\u591a\u6837\u5316\u4e14\u5177\u6709\u4ee3\u8868\u6027\u3002<\/p>\n<\/li>\n

          2. \n

            \u5e76\u884c\u5904\u7406<\/strong>\uff1a\u5728\u5927\u578b\u6570\u636e\u96c6\u4e0a\u8bad\u7ec3 VQGAN \u53ef\u80fd\u9700\u8981\u5927\u91cf\u8ba1\u7b97\u3002\u4ee3\u7406\u670d\u52a1\u5668\u53ef\u4ee5\u5c06\u5de5\u4f5c\u8d1f\u8f7d\u5206\u914d\u5230\u591a\u53f0\u673a\u5668\u4e0a\uff0c\u4ece\u800c\u52a0\u5feb\u8bad\u7ec3\u8fc7\u7a0b\u3002<\/p>\n<\/li>\n

          3. \n

            API \u7aef\u70b9<\/strong>\uff1a\u4ee3\u7406\u670d\u52a1\u5668\u53ef\u4ee5\u4f5c\u4e3a\u90e8\u7f72 VQGAN \u6a21\u578b\u7684 API \u7aef\u70b9\uff0c\u4f7f\u7528\u6237\u80fd\u591f\u8fdc\u7a0b\u4e0e\u6a21\u578b\u4ea4\u4e92\u5e76\u6309\u9700\u751f\u6210\u56fe\u50cf\u3002<\/p>\n<\/li>\n<\/ol>\n

            \u76f8\u5173\u94fe\u63a5<\/h2>\n

            \u6709\u5173\u77e2\u91cf\u91cf\u5316\u751f\u6210\u5bf9\u6297\u7f51\u7edc (VQGAN) \u53ca\u5176\u76f8\u5173\u4e3b\u9898\u7684\u66f4\u591a\u4fe1\u606f\uff0c\u8bf7\u53c2\u9605\u4ee5\u4e0b\u8d44\u6e90\uff1a<\/p>\n

              \n
            1. \n

              DeepMind \u535a\u5ba2 \u2013 \u4ecb\u7ecd VQ-VAE-2<\/a><\/p>\n<\/li>\n

            2. \n

              arXiv - VQ-VAE-2\uff1a\u6539\u8fdb\u7684 GAN \u548c VAE \u79bb\u6563\u9690\u53d8\u91cf\u8bad\u7ec3<\/a><\/p>\n<\/li>\n

            3. \n

              GitHub \u2013 VQ-VAE-2 \u5b9e\u73b0<\/a><\/p>\n<\/li>\n

            4. \n

              OpenAI \u2013 CLIP\uff1a\u8fde\u63a5\u6587\u672c\u548c\u56fe\u50cf<\/a><\/p>\n<\/li>\n

            5. \n

              arXiv \u2013 CLIP\uff1a\u5927\u89c4\u6a21\u8fde\u63a5\u6587\u672c\u548c\u56fe\u50cf<\/a><\/p>\n<\/li>\n<\/ol>\n

              \u901a\u8fc7\u63a2\u7d22\u8fd9\u4e9b\u8d44\u6e90\uff0c\u60a8\u53ef\u4ee5\u66f4\u6df1\u5165\u5730\u4e86\u89e3\u77e2\u91cf\u91cf\u5316\u751f\u6210\u5bf9\u6297\u7f51\u7edc (VQGAN) \u53ca\u5176\u5728\u4eba\u5de5\u667a\u80fd\u548c\u521b\u610f\u5185\u5bb9\u751f\u6210\u9886\u57df\u7684\u5e94\u7528\u3002<\/p>","protected":false},"featured_media":470817,"menu_order":0,"template":"","meta":{"_acf_changed":false,"content-type":"","inline_featured_image":false,"footnotes":""},"class_list":["post-479505","wiki","type-wiki","status-publish","has-post-thumbnail","hentry"],"acf":{"faq_title":"Frequently Asked Questions about Vector Quantized Generative Adversarial Network (VQGAN)<\/mark>","faq_items":[{"question":"What is Vector Quantized Generative Adversarial Network (VQGAN)?","answer":"

              Vector Quantized Generative Adversarial Network (VQGAN) is an advanced deep learning model that combines Generative Adversarial Networks (GANs) and Vector Quantization (VQ) techniques. It excels in generating high-quality images and offers improved control over the creative content generation process.<\/p>"},{"question":"How does VQGAN work?","answer":"

              VQGAN consists of a generator and a discriminator, similar to traditional GANs. The key innovation lies in its encoder architecture, which maps input images to discrete latent codes. These codes are then quantized using a predefined set of embeddings in a codebook. The model is trained to minimize reconstruction and adversarial losses, resulting in realistic and visually appealing image synthesis.<\/p>"},{"question":"What are the key features of VQGAN?","answer":"