resnet as student network and CLIP as teacher network

[gqingc]

Is it feasible to use resnet as student network and CLIP as teacher network? Their output tensors are different. Is it feasible for me to reshape? My code is as follows: z = self.encoder_q(img) //encode is resnet network print("z0的shape=") print(z.shape)# //[16,128] ,16 is batchsize z=z.unsqueeze(2) print("z1的shape=") print(z.shape) //[16,128,1] z=self.decoder(z) //一个卷积层， print("z2的shape=") print(z.shape) //[16,38400,1] z=z.view(8, 50, 768) //Reshape is used here

x_rec = self.decoder(z)

    # print("x_rec的shape=")
    # print(x_rec.shape)
    self.feature_model.eval()
    with torch.no_grad():
        x = normalize_clip(unnormalize(im1))
        print("x0的shape=")
        print(x.shape)
        # x = self.resize_func(x)
        # print("x0的shape=")
        # print(x.shape)
        x_tgt = self.feature_model.encode_image_featuremap(x)
        print("x1的shape=")
        print(x_tgt.shape)
        x_tgt = self.feature_model.visual.ln_post(x_tgt)
        x_tgt = x_tgt.detach()
        x_tgt = self.ln_tgt(x_tgt)
        print("x2的shape=")
        print(x_tgt.shape)   //[16,50,768]

    loss_FD = self.loss_feat(z, x_tgt)
    loss1 = loss_FD.mean()