bring in attention-based upsampling to strengthen vqgan-vae, seems to work as advertised in initial experiments in GAN

README.md

@@ -12,7 +12,7 @@ This model is SOTA for text-to-image for now.
 
 Please join <a href="https://discord.gg/xBPBXfcFHd"><img alt="Join us on Discord" src="https://img.shields.io/discord/823813159592001537?color=5865F2&logo=discord&logoColor=white"></a> if you are interested in helping out with the replication
 
-There was enough interest for a Jax version. It will be completed after the Pytorch version shows signs of life on my toy tasks. <a href="https://github.com/lucidrains/dalle2-jax">Placeholder repository</a>. I will also eventually extend this to <a href="https://github.com/lucidrains/dalle2-video">text to video</a>, once the repository is in a good place.
+There was enough interest for a <a href="https://github.com/lucidrains/dalle2-jax">Jax version</a>. I will also eventually extend this to <a href="https://github.com/lucidrains/dalle2-video">text to video</a>, once the repository is in a good place.
 
 ## Install
 
@@ -246,13 +246,6 @@ loss = decoder(images, unet_number = 2)
 loss.backward()
 
 # do the above for many steps for both unets
-
-# then it will learn to generate images based on the CLIP image embeddings
-
-# chaining the unets from lowest resolution to highest resolution (thus cascading)
-
-mock_image_embed = torch.randn(1, 512).cuda()
-images = decoder.sample(mock_image_embed) # (1, 3, 512, 512)
 ```
 
 Finally, to generate the DALL-E2 images from text. Insert the trained `DiffusionPrior` as well as the `Decoder` (which wraps `CLIP`, the causal transformer, and unet(s))
@@ -533,10 +526,12 @@ Once built, images will be saved to the same directory the command is invoked
 - [ ] spend one day cleaning up tech debt in decoder
 - [ ] become an expert with unets, cleanup unet code, make it fully configurable, port all learnings over to https://github.com/lucidrains/x-unet
 - [ ] copy the cascading ddpm code to a separate repo (perhaps https://github.com/lucidrains/denoising-diffusion-pytorch) as the main contribution of dalle2 really is just the prior network
+- [ ] transcribe code to Jax, which lowers the activation energy for distributed training, given access to TPUs
 - [ ] train on a toy task, offer in colab
 - [ ] extend diffusion head to use diffusion-gan (potentially using lightweight-gan) to speed up inference
 - [ ] bring in tools to train vqgan-vae
 - [ ] bring in vit-vqgan https://arxiv.org/abs/2110.04627 for the latent diffusion
+- [ ] experiment with https://arxiv.org/abs/2112.11435 as upsampler, test in https://github.com/lucidrains/lightweight-gan first
 
 ## Citations
 
@@ -568,7 +563,7 @@ Once built, images will be saved to the same directory the command is invoked
 
 ```bibtex
 @inproceedings{Liu2022ACF,
-    title   = {A ConvNet for the 2020s},
+    title   = {A ConvNet for the 2020https://arxiv.org/abs/2112.11435s},
     author  = {Zhuang Liu and Hanzi Mao and Chaozheng Wu and Christoph Feichtenhofer and Trevor Darrell and Saining Xie},
     year    = {2022}
 }
@@ -582,4 +577,14 @@ Once built, images will be saved to the same directory the command is invoked
 }
 ```
 
+```bibtex
+@article{Arar2021LearnedQF,
+    title   = {Learned Queries for Efficient Local Attention},
+    author  = {Moab Arar and Ariel Shamir and Amit H. Bermano},
+    journal = {ArXiv},
+    year    = {2021},
+    volume  = {abs/2112.11435}
+}
+```
+
 *Creating noise from data is easy; creating data from noise is generative modeling.* - Yang Song's <a href="https://arxiv.org/abs/2011.13456">paper</a>

dalle2_pytorch/dalle2_pytorch.py

@@ -483,7 +483,7 @@ class DiffusionPrior(nn.Module):
         timesteps = 1000,
         cond_drop_prob = 0.2,
         loss_type = "l1",
-        predict_x0 = True,
+        predict_x_start = True,
         beta_schedule = "cosine",
     ):
         super().__init__()
@@ -497,7 +497,7 @@ class DiffusionPrior(nn.Module):
         self.image_size = clip.image_size
         self.cond_drop_prob = cond_drop_prob
 
-        self.predict_x0 = predict_x0
+        self.predict_x_start = predict_x_start
         # in paper, they do not predict the noise, but predict x0 directly for image embedding, claiming empirically better results. I'll just offer both.
 
         if beta_schedule == "cosine":
@@ -586,14 +586,14 @@ class DiffusionPrior(nn.Module):
     def p_mean_variance(self, x, t, text_cond, clip_denoised: bool):
         pred = self.net(x, t, **text_cond)
 
-        if self.predict_x0:
+        if self.predict_x_start:
             x_recon = pred
             # not 100% sure of this above line - for any spectators, let me know in the github issues (or through a pull request) if you know how to correctly do this
             # i'll be rereading https://arxiv.org/abs/2111.14822, where i think a similar approach is taken
         else:
             x_recon = self.predict_start_from_noise(x, t = t, noise = pred)
 
-        if clip_denoised and not self.predict_x0:
+        if clip_denoised and not self.predict_x_start:
             x_recon.clamp_(-1., 1.)
 
         model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t)
@@ -639,7 +639,7 @@ class DiffusionPrior(nn.Module):
             **text_cond
         )
 
-        to_predict = noise if not self.predict_x0 else image_embed
+        to_predict = noise if not self.predict_x_start else image_embed
 
         if self.loss_type == 'l1':
             loss = F.l1_loss(to_predict, x_recon)
@@ -1121,7 +1121,8 @@ class Decoder(nn.Module):
         cond_drop_prob = 0.2,
         loss_type = 'l1',
         beta_schedule = 'cosine',
-        predict_x0 = False,
+        predict_x_start = False,
+        predict_x_start_for_latent_diffusion = False,
         image_sizes = None,                         # for cascading ddpm, image size at each stage
         lowres_cond_upsample_mode = 'bilinear',     # cascading ddpm - low resolution upsample mode
         lowres_downsample_first = True,             # cascading ddpm - resizes to lower resolution, then to next conditional resolution + blur
@@ -1172,7 +1173,7 @@ class Decoder(nn.Module):
 
         # predict x0 config
 
-        self.predict_x0 = cast_tuple(predict_x0, len(unets))
+        self.predict_x_start = cast_tuple(predict_x_start, len(unets)) if not predict_x_start_for_latent_diffusion else tuple(map(lambda t: isinstance(t, VQGanVAE), self.vaes))
 
         # cascading ddpm related stuff
 
@@ -1292,31 +1293,31 @@ class Decoder(nn.Module):
         posterior_log_variance_clipped = extract(self.posterior_log_variance_clipped, t, x_t.shape)
         return posterior_mean, posterior_variance, posterior_log_variance_clipped
 
-    def p_mean_variance(self, unet, x, t, image_embed, text_encodings = None, lowres_cond_img = None, clip_denoised = True, predict_x0 = False, cond_scale = 1.):
+    def p_mean_variance(self, unet, x, t, image_embed, text_encodings = None, lowres_cond_img = None, clip_denoised = True, predict_x_start = False, cond_scale = 1.):
         pred = unet.forward_with_cond_scale(x, t, image_embed = image_embed, text_encodings = text_encodings, cond_scale = cond_scale, lowres_cond_img = lowres_cond_img)
 
-        if predict_x0:
+        if predict_x_start:
             x_recon = pred
         else:
             x_recon = self.predict_start_from_noise(x, t = t, noise = pred)
 
-        if clip_denoised and not predict_x0:
+        if clip_denoised and not predict_x_start:
             x_recon.clamp_(-1., 1.)
 
         model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t)
         return model_mean, posterior_variance, posterior_log_variance
 
     @torch.no_grad()
-    def p_sample(self, unet, x, t, image_embed, text_encodings = None, cond_scale = 1., lowres_cond_img = None, predict_x0 = False, clip_denoised = True, repeat_noise = False):
+    def p_sample(self, unet, x, t, image_embed, text_encodings = None, cond_scale = 1., lowres_cond_img = None, predict_x_start = False, clip_denoised = True, repeat_noise = False):
         b, *_, device = *x.shape, x.device
-        model_mean, _, model_log_variance = self.p_mean_variance(unet, x = x, t = t, image_embed = image_embed, text_encodings = text_encodings, cond_scale = cond_scale, lowres_cond_img = lowres_cond_img, clip_denoised = clip_denoised, predict_x0 = predict_x0)
+        model_mean, _, model_log_variance = self.p_mean_variance(unet, x = x, t = t, image_embed = image_embed, text_encodings = text_encodings, cond_scale = cond_scale, lowres_cond_img = lowres_cond_img, clip_denoised = clip_denoised, predict_x_start = predict_x_start)
         noise = noise_like(x.shape, device, repeat_noise)
         # no noise when t == 0
         nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1)))
         return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise
 
     @torch.no_grad()
-    def p_sample_loop(self, unet, shape, image_embed, predict_x0 = False, lowres_cond_img = None, text_encodings = None, cond_scale = 1):
+    def p_sample_loop(self, unet, shape, image_embed, predict_x_start = False, lowres_cond_img = None, text_encodings = None, cond_scale = 1):
         device = self.betas.device
 
         b = shape[0]
@@ -1331,7 +1332,7 @@ class Decoder(nn.Module):
                 text_encodings = text_encodings,
                 cond_scale = cond_scale,
                 lowres_cond_img = lowres_cond_img,
-                predict_x0 = predict_x0
+                predict_x_start = predict_x_start
             )
 
         return img
@@ -1344,7 +1345,7 @@ class Decoder(nn.Module):
             extract(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise
         )
 
-    def p_losses(self, unet, x_start, t, *, image_embed, lowres_cond_img = None, text_encodings = None, predict_x0 = False, noise = None):
+    def p_losses(self, unet, x_start, t, *, image_embed, lowres_cond_img = None, text_encodings = None, predict_x_start = False, noise = None):
         noise = default(noise, lambda: torch.randn_like(x_start))
 
         x_noisy = self.q_sample(x_start = x_start, t = t, noise = noise)
@@ -1358,7 +1359,7 @@ class Decoder(nn.Module):
             cond_drop_prob = self.cond_drop_prob
         )
 
-        target = noise if not predict_x0 else x_start
+        target = noise if not predict_x_start else x_start
 
         if self.loss_type == 'l1':
             loss = F.l1_loss(target, x_recon)
@@ -1380,7 +1381,7 @@ class Decoder(nn.Module):
 
         img = None
 
-        for unet, vae, channel, image_size, predict_x0 in tqdm(zip(self.unets, self.vaes, self.sample_channels, self.image_sizes, self.predict_x0)):
+        for unet, vae, channel, image_size, predict_x_start in tqdm(zip(self.unets, self.vaes, self.sample_channels, self.image_sizes, self.predict_x_start)):
             with self.one_unet_in_gpu(unet = unet):
                 lowres_cond_img = None
                 shape = (batch_size, channel, image_size, image_size)
@@ -1400,7 +1401,7 @@ class Decoder(nn.Module):
                     image_embed = image_embed,
                     text_encodings = text_encodings,
                     cond_scale = cond_scale,
-                    predict_x0 = predict_x0,
+                    predict_x_start = predict_x_start,
                     lowres_cond_img = lowres_cond_img
                 )
 
@@ -1424,7 +1425,7 @@ class Decoder(nn.Module):
 
         target_image_size = self.image_sizes[unet_index]
         vae = self.vaes[unet_index]
-        predict_x0 = self.predict_x0[unet_index]
+        predict_x_start = self.predict_x_start[unet_index]
 
         b, c, h, w, device, = *image.shape, image.device
 
@@ -1448,7 +1449,7 @@ class Decoder(nn.Module):
             if exists(lowres_cond_img):
                 lowres_cond_img = vae.encode(lowres_cond_img)
 
-        return self.p_losses(unet, image, times, image_embed = image_embed, text_encodings = text_encodings, lowres_cond_img = lowres_cond_img, predict_x0 = predict_x0)
+        return self.p_losses(unet, image, times, image_embed = image_embed, text_encodings = text_encodings, lowres_cond_img = lowres_cond_img, predict_x_start = predict_x_start)
 
 # main class
 

dalle2_pytorch/vqgan_vae.py

@@ -243,6 +243,112 @@ class ResBlock(nn.Module):
     def forward(self, x):
         return self.net(x) + x
 
+# attention-based upsampling
+# from https://arxiv.org/abs/2112.11435
+
+class QueryAndAttend(nn.Module):
+    def __init__(
+        self,
+        *,
+        dim,
+        num_queries = 1,
+        dim_head = 32,
+        heads = 8,
+        window_size = 3
+    ):
+        super().__init__()
+        self.scale = dim_head ** -0.5
+        inner_dim = dim_head * heads
+        self.heads = heads
+        self.dim_head = dim_head
+        self.window_size = window_size
+        self.num_queries = num_queries
+
+        self.rel_pos_bias = nn.Parameter(torch.randn(heads, num_queries, window_size * window_size, 1, 1))
+
+        self.queries = nn.Parameter(torch.randn(heads, num_queries, dim_head))
+        self.to_kv = nn.Conv2d(dim, dim_head * 2, 1, bias = False)
+        self.to_out = nn.Conv2d(inner_dim, dim, 1, bias = False)
+
+    def forward(self, x):
+        """
+        einstein notation
+        b - batch
+        h - heads
+        l - num queries
+        d - head dimension
+        x - height
+        y - width
+        j - source sequence for attending to (kernel size squared in this case)
+        """
+
+        wsz, heads, dim_head, num_queries = self.window_size, self.heads, self.dim_head, self.num_queries
+        batch, _, height, width = x.shape
+
+        is_one_query = self.num_queries == 1
+
+        # queries, keys, values
+
+        q = self.queries * self.scale
+        k, v = self.to_kv(x).chunk(2, dim = 1)
+
+        # similarities
+
+        sim = einsum('h l d, b d x y -> b h l x y', q, k)
+        sim = rearrange(sim, 'b ... x y -> b (...) x y')
+
+        # unfold the similarity scores, with float(-inf) as padding value
+
+        mask_value = -torch.finfo(sim.dtype).max
+        sim = F.pad(sim, ((wsz // 2,) * 4), value = mask_value)
+        sim = F.unfold(sim, kernel_size = wsz)
+        sim = rearrange(sim, 'b (h l j) (x y) -> b h l j x y', h = heads, l = num_queries, x = height, y = width)
+
+        # rel pos bias
+
+        sim = sim + self.rel_pos_bias
+
+        # numerically stable attention
+
+        sim = sim - sim.amax(dim = -3, keepdim = True).detach()
+        attn = sim.softmax(dim = -3)
+
+        # unfold values
+
+        v = F.pad(v, ((wsz // 2,) * 4), value = 0.)
+        v = F.unfold(v, kernel_size = wsz)
+        v = rearrange(v, 'b (d j) (x y) -> b d j x y', d = dim_head, x = height, y = width)
+
+        # aggregate values
+
+        out = einsum('b h l j x y, b d j x y -> b l h d x y', attn, v)
+
+        # combine heads
+
+        out = rearrange(out, 'b l h d x y -> (b l) (h d) x y')
+        out = self.to_out(out)
+        out = rearrange(out, '(b l) d x y -> b l d x y', b = batch)
+
+        # return original input if one query
+
+        if is_one_query:
+            out = rearrange(out, 'b 1 ... -> b ...')
+
+        return out
+
+class QueryAttnUpsample(nn.Module):
+    def __init__(self, dim, **kwargs):
+        super().__init__()
+        self.norm = LayerNormChan(dim)
+        self.qna = QueryAndAttend(dim = dim, num_queries = 4, **kwargs)
+
+    def forward(self, x):
+        x = self.norm(x)
+        out = self.qna(x)
+        out = rearrange(out, 'b (w1 w2) c h w -> b c (h w1) (w w2)', w1 = 2, w2 = 2)
+        return out
+
+# vqgan attention layer
 class VQGanAttention(nn.Module):
     def __init__(
         self,
@@ -375,7 +481,7 @@ class VQGanVAE(nn.Module):
 
         for layer_index, (dim_in, dim_out), layer_num_resnet_blocks, layer_use_attn in zip(range(layers), dim_pairs, num_resnet_blocks, use_attn):
             append(self.encoders, nn.Sequential(nn.Conv2d(dim_in, dim_out, 4, stride = 2, padding = 1), leaky_relu()))
-            prepend(self.decoders, nn.Sequential(nn.Upsample(scale_factor = 2, mode = 'bilinear', align_corners = False), nn.Conv2d(dim_out, dim_in, 3, padding = 1), leaky_relu()))
+            prepend(self.decoders, nn.Sequential(QueryAttnUpsample(dim_out), nn.Conv2d(dim_out, dim_in, 3, padding = 1), leaky_relu()))
 
             if layer_use_attn:
                 prepend(self.decoders, VQGanAttention(dim = dim_out, heads = attn_heads, dim_head = attn_dim_head, dropout = attn_dropout))

setup.py

@@ -10,7 +10,7 @@ setup(
       'dream = dalle2_pytorch.cli:dream'
     ],
   },
-  version = '0.0.41',
+  version = '0.0.44',
   license='MIT',
   description = 'DALL-E 2',
   author = 'Phil Wang',