Merge pull request #75 from lucidrains/fsq

Fsq

Author: lucidrains

README.md

@@ -251,6 +251,40 @@ indices = quantizer(x) # (1, 1024, 16) - (batch, seq, num_codebooks)
 
 This repository should also automatically synchronizing the codebooks in a multi-process setting. If somehow it isn't, please open an issue. You can override whether to synchronize codebooks or not by setting `sync_codebook = True | False`
 
+### Finite Scalar Quantization
+
+<img src="./fsq.png" width="500px"></img>
+
+|                  | VQ | FSQ |
+|------------------|----|-----|
+| Quantization     | argmin_c \|\| z-c \|\| | round(f(z)) |
+| Gradients        | Straight Through Estimation (STE) | STE |
+| Auxiliary Losses | Commitment, codebook, entropy loss, ... | N/A |
+| Tricks           | EMA on codebook, codebook splitting, projections, ...| N/A |
+| Parameters       | Codebook | N/A |
+
+[This](https://arxiv.org/abs/2309.15505) work out of Google Deepmind aims to vastly simplify the way vector quantization is done for generative modeling, removing the need for commitment losses, EMA updating of the codebook, as well as tackle the issues with codebook collapse or insufficient utilization. They simply round each scalar into discrete levels with straight through gradients; the codes become uniform points in a hypercube.
+
+
+```python
+import torch
+from vector_quantize_pytorch import FSQ
+
+levels = [8,5,5,5] # see 4.1 and A.4.1 in the paper
+quantizer = FSQ(levels)
+
+x = torch.randn(1, 1024, quantizer.dim)
+xhat, indices = quantizer(x)
+
+print(xhat.shape)    # (1, 1024, 4) - (batch, seq, dim)
+print(indices.shape) # (1, 1024)    - (batch, seq)
+
+assert torch.all(xhat == quantizer.indices_to_codes(indices))
+assert torch.all(xhat == quantizer.implicit_codebook[indices])
+```
+
+
+
 ## Todo
 
 - [x] allow for multi-headed codebooks
@@ -384,3 +418,14 @@ This repository should also automatically synchronizing the codebooks in a multi
     primaryClass = {cs.CV}
 }
 ```
+
+```bibtex
+@misc{mentzer2023finite,
+    title   = {Finite Scalar Quantization: VQ-VAE Made Simple},
+    author  = {Fabian Mentzer and David Minnen and Eirikur Agustsson and Michael Tschannen},
+    year    = {2023},
+    eprint  = {2309.15505},
+    archivePrefix = {arXiv},
+    primaryClass = {cs.CV}
+}
+```

examples/autoencoder.py

@@ -41,7 +41,7 @@ def __init__(self, **vq_kwargs):
     def forward(self, x):
         for layer in self.layers:
             if isinstance(layer, VectorQuantize):
-                x_flat, indices, commit_loss = layer(x)
+                x, indices, commit_loss = layer(x)
             else:
                 x = layer(x)
 

examples/autoencoder_fsq.py

@@ -0,0 +1,94 @@
+# FashionMnist VQ experiment with various settings, using FSQ.
+# From https://github.com/minyoungg/vqtorch/blob/main/examples/autoencoder.py
+
+from tqdm.auto import trange
+
+import math
+import torch
+import torch.nn as nn
+from torchvision import datasets, transforms
+from torch.utils.data import DataLoader
+
+from vector_quantize_pytorch import FSQ
+
+
+lr = 3e-4
+train_iter = 1000
+levels = [8, 6, 5] # target size 2^8, actual size 240
+num_codes = math.prod(levels)
+seed = 1234
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+
+class SimpleFSQAutoEncoder(nn.Module):
+    def __init__(self, levels: list[int]):
+        super().__init__()
+        self.layers = nn.ModuleList(
+            [
+                nn.Conv2d(1, 16, kernel_size=3, stride=1, padding=1),
+                nn.MaxPool2d(kernel_size=2, stride=2),
+                nn.GELU(),
+                nn.Conv2d(16, 8, kernel_size=3, stride=1, padding=1),
+                nn.Conv2d(8, 8, kernel_size=6, stride=3, padding=0),
+                FSQ(levels),
+                nn.ConvTranspose2d(8, 8, kernel_size=6, stride=3, padding=0),
+                nn.Conv2d(8, 16, kernel_size=4, stride=1, padding=2),
+                nn.GELU(),
+                nn.Upsample(scale_factor=2, mode="nearest"),
+                nn.Conv2d(16, 1, kernel_size=3, stride=1, padding=2),
+            ]
+        )
+        return
+
+    def forward(self, x):
+        for layer in self.layers:
+            if isinstance(layer, FSQ):
+                x, indices = layer(x)
+            else:
+                x = layer(x)
+
+        return x.clamp(-1, 1), indices
+
+
+def train(model, train_loader, train_iterations=1000):
+    def iterate_dataset(data_loader):
+        data_iter = iter(data_loader)
+        while True:
+            try:
+                x, y = next(data_iter)
+            except StopIteration:
+                data_iter = iter(data_loader)
+                x, y = next(data_iter)
+            yield x.to(device), y.to(device)
+
+    for _ in (pbar := trange(train_iterations)):
+        opt.zero_grad()
+        x, _ = next(iterate_dataset(train_loader))
+        out, indices = model(x)
+        rec_loss = (out - x).abs().mean()
+        rec_loss.backward()
+
+        opt.step()
+        pbar.set_description(
+            f"rec loss: {rec_loss.item():.3f} | "
+            + f"active %: {indices.unique().numel() / num_codes * 100:.3f}"
+        )
+    return
+
+
+transform = transforms.Compose(
+    [transforms.ToTensor(), transforms.Normalize((0.5,), (0.5,))]
+)
+train_dataset = DataLoader(
+    datasets.FashionMNIST(
+        root="~/data/fashion_mnist", train=True, download=True, transform=transform
+    ),
+    batch_size=256,
+    shuffle=True,
+)
+
+print("baseline")
+torch.random.manual_seed(seed)
+model = SimpleFSQAutoEncoder(levels).to(device)
+opt = torch.optim.AdamW(model.parameters(), lr=lr)
+train(model, train_dataset, train_iterations=train_iter)

fsq.png

@@ -3,7 +3,7 @@
 setup(
   name = 'vector_quantize_pytorch',
   packages = find_packages(),
-  version = '1.7.1',
+  version = '1.8.0',
   license='MIT',
   description = 'Vector Quantization - Pytorch',
   long_description_content_type = 'text/markdown',

setup.py

@@ -1,3 +1,4 @@
 from vector_quantize_pytorch.vector_quantize_pytorch import VectorQuantize
 from vector_quantize_pytorch.residual_vq import ResidualVQ, GroupedResidualVQ
-from vector_quantize_pytorch.random_projection_quantizer import RandomProjectionQuantizer
+from vector_quantize_pytorch.random_projection_quantizer import RandomProjectionQuantizer
+from vector_quantize_pytorch.finite_scalar_quantization import FSQ

vector_quantize_pytorch/__init__.py

@@ -0,0 +1,66 @@
+"""
+Finite Scalar Quantization: VQ-VAE Made Simple - https://arxiv.org/abs/2309.15505
+Code adapted from Jax version in Appendix A.1
+"""
+
+import torch
+import torch.nn as nn
+
+
+def round_ste(z: torch.Tensor) -> torch.Tensor:
+    """Round with straight through gradients."""
+    zhat = z.round()
+    return z + (zhat - z).detach()
+
+
+class FSQ(nn.Module):
+    def __init__(self, levels: list[int]):
+        super().__init__()
+        _levels = torch.tensor(levels, dtype=torch.int32)
+        self.register_buffer("_levels", _levels)
+
+        _basis = torch.cumprod(torch.tensor([1] + levels[:-1]), dim=0, dtype=torch.int32)
+        self.register_buffer("_basis", _basis)
+
+        self.dim = len(levels)
+        self.n_codes = self._levels.prod().item()
+        implicit_codebook = self.indices_to_codes(torch.arange(self.n_codes))
+        self.register_buffer("implicit_codebook", implicit_codebook)
+
+    def forward(self, z: torch.Tensor) -> torch.Tensor:
+        zhat = self.quantize(z)
+        indices = self.codes_to_indices(zhat)
+        return zhat, indices
+
+    def bound(self, z: torch.Tensor, eps: float = 1e-3) -> torch.Tensor:
+        """Bound `z`, an array of shape (..., d)."""
+        half_l = (self._levels - 1) * (1 - eps) / 2
+        offset = torch.where(self._levels % 2 == 0, 0.5, 0.0)
+        shift = (offset / half_l).tan()
+        return (z + shift).tanh() * half_l - offset
+
+    def quantize(self, z: torch.Tensor) -> torch.Tensor:
+        """Quanitzes z, returns quantized zhat, same shape as z."""
+        quantized = round_ste(self.bound(z))
+        half_width = self._levels // 2 # Renormalize to [-1, 1].
+        return quantized / half_width
+    
+    def _scale_and_shift(self, zhat_normalized: torch.Tensor) -> torch.Tensor:
+        half_width = self._levels // 2
+        return (zhat_normalized * half_width) + half_width
+    
+    def _scale_and_shift_inverse(self, zhat: torch.Tensor) -> torch.Tensor:
+        half_width = self._levels // 2
+        return (zhat - half_width) / half_width
+    
+    def codes_to_indices(self, zhat: torch.Tensor) -> torch.Tensor:
+        """Converts a `code` to an index in the codebook."""
+        assert zhat.shape[-1] == self.dim
+        zhat = self._scale_and_shift(zhat)
+        return (zhat * self._basis).sum(dim=-1).to(torch.int32)
+    
+    def indices_to_codes(self, indices: torch.Tensor) -> torch.Tensor:
+        """Inverse of `codes_to_indices`."""
+        indices = indices.unsqueeze(-1)
+        codes_non_centered = (indices // self._basis) % self._levels
+        return self._scale_and_shift_inverse(codes_non_centered)