check_ckpt.py

# %pip install einops==0.3.0
import importlib
import albumentations

import torch
import torch.nn.functional as F
from taming.data.utils import custom_collate
# pytorch_diffusion + derived encoder decoder
import math
import torch
import torch.nn as nn
import numpy as np
from einops import rearrange

def get_timestep_embedding(timesteps, embedding_dim):
    """
    This matches the implementation in Denoising Diffusion Probabilistic Models:
    From Fairseq.
    Build sinusoidal embeddings.
    This matches the implementation in tensor2tensor, but differs slightly
    from the description in Section 3.5 of "Attention Is All You Need".
    """
    assert len(timesteps.shape) == 1

    half_dim = embedding_dim // 2
    emb = math.log(10000) / (half_dim - 1)
    emb = torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb)
    emb = emb.to(device=timesteps.device)
    emb = timesteps.float()[:, None] * emb[None, :]
    emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1)
    if embedding_dim % 2 == 1:  # zero pad
        emb = torch.nn.functional.pad(emb, (0,1,0,0))
    return emb


def nonlinearity(x):
    # swish
    return x*torch.sigmoid(x)


def Normalize(in_channels):
    return torch.nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True)


class Upsample(nn.Module):
    def __init__(self, in_channels, with_conv):
        super().__init__()
        self.with_conv = with_conv
        if self.with_conv:
            self.conv = torch.nn.Conv2d(in_channels,
                                        in_channels,
                                        kernel_size=3,
                                        stride=1,
                                        padding=1)

    def forward(self, x):
        x = torch.nn.functional.interpolate(x, scale_factor=2.0, mode="nearest")
        if self.with_conv:
            x = self.conv(x)
        return x


class Downsample(nn.Module):
    def __init__(self, in_channels, with_conv):
        super().__init__()
        self.with_conv = with_conv
        if self.with_conv:
            # no asymmetric padding in torch conv, must do it ourselves
            self.conv = torch.nn.Conv2d(in_channels,
                                        in_channels,
                                        kernel_size=3,
                                        stride=2,
                                        padding=0)

    def forward(self, x):
        if self.with_conv:
            pad = (0,1,0,1)
            x = torch.nn.functional.pad(x, pad, mode="constant", value=0)
            x = self.conv(x)
        else:
            x = torch.nn.functional.avg_pool2d(x, kernel_size=2, stride=2)
        return x


class ResnetBlock(nn.Module):
    def __init__(self, *, in_channels, out_channels=None, conv_shortcut=False,
                 dropout, temb_channels=512):
        super().__init__()
        self.in_channels = in_channels
        out_channels = in_channels if out_channels is None else out_channels
        self.out_channels = out_channels
        self.use_conv_shortcut = conv_shortcut

        self.norm1 = Normalize(in_channels)
        self.conv1 = torch.nn.Conv2d(in_channels,
                                     out_channels,
                                     kernel_size=3,
                                     stride=1,
                                     padding=1)
        if temb_channels > 0:
            self.temb_proj = torch.nn.Linear(temb_channels,
                                             out_channels)
        self.norm2 = Normalize(out_channels)
        self.dropout = torch.nn.Dropout(dropout)
        self.conv2 = torch.nn.Conv2d(out_channels,
                                     out_channels,
                                     kernel_size=3,
                                     stride=1,
                                     padding=1)
        if self.in_channels != self.out_channels:
            if self.use_conv_shortcut:
                self.conv_shortcut = torch.nn.Conv2d(in_channels,
                                                     out_channels,
                                                     kernel_size=3,
                                                     stride=1,
                                                     padding=1)
            else:
                self.nin_shortcut = torch.nn.Conv2d(in_channels,
                                                    out_channels,
                                                    kernel_size=1,
                                                    stride=1,
                                                    padding=0)

    def forward(self, x, temb):
        h = x
        h = self.norm1(h)
        h = nonlinearity(h)
        h = self.conv1(h)

        if temb is not None:
            h = h + self.temb_proj(nonlinearity(temb))[:,:,None,None]

        h = self.norm2(h)
        h = nonlinearity(h)
        h = self.dropout(h)
        h = self.conv2(h)

        if self.in_channels != self.out_channels:
            if self.use_conv_shortcut:
                x = self.conv_shortcut(x)
            else:
                x = self.nin_shortcut(x)

        return x+h


class AttnBlock(nn.Module):
    def __init__(self, in_channels):
        super().__init__()
        self.in_channels = in_channels

        self.norm = Normalize(in_channels)
        self.q = torch.nn.Conv2d(in_channels,
                                 in_channels,
                                 kernel_size=1,
                                 stride=1,
                                 padding=0)
        self.k = torch.nn.Conv2d(in_channels,
                                 in_channels,
                                 kernel_size=1,
                                 stride=1,
                                 padding=0)
        self.v = torch.nn.Conv2d(in_channels,
                                 in_channels,
                                 kernel_size=1,
                                 stride=1,
                                 padding=0)
        self.proj_out = torch.nn.Conv2d(in_channels,
                                        in_channels,
                                        kernel_size=1,
                                        stride=1,
                                        padding=0)


    def forward(self, x):
        h_ = x
        h_ = self.norm(h_)
        q = self.q(h_)
        k = self.k(h_)
        v = self.v(h_)

        # compute attention
        b,c,h,w = q.shape
        q = q.reshape(b,c,h*w)
        q = q.permute(0,2,1)   # b,hw,c
        k = k.reshape(b,c,h*w) # b,c,hw
        w_ = torch.bmm(q,k)     # b,hw,hw    w[b,i,j]=sum_c q[b,i,c]k[b,c,j]
        w_ = w_ * (int(c)**(-0.5))
        w_ = torch.nn.functional.softmax(w_, dim=2)

        # attend to values
        v = v.reshape(b,c,h*w)
        w_ = w_.permute(0,2,1)   # b,hw,hw (first hw of k, second of q)
        h_ = torch.bmm(v,w_)     # b, c,hw (hw of q) h_[b,c,j] = sum_i v[b,c,i] w_[b,i,j]
        h_ = h_.reshape(b,c,h,w)

        h_ = self.proj_out(h_)

        return x+h_


class Encoder(nn.Module):
    def __init__(self, *, ch, out_ch, ch_mult=(1,2,4,8), num_res_blocks,
                 attn_resolutions, dropout=0.0, resamp_with_conv=True, in_channels,
                 resolution, z_channels, double_z=True, **ignore_kwargs):
        super().__init__()
        self.ch = ch
        self.temb_ch = 0
        self.num_resolutions = len(ch_mult)
        self.num_res_blocks = num_res_blocks
        self.resolution = resolution
        self.in_channels = in_channels

        # downsampling
        self.conv_in = torch.nn.Conv2d(in_channels,
                                       self.ch,
                                       kernel_size=3,
                                       stride=1,
                                       padding=1)

        curr_res = resolution
        in_ch_mult = (1,)+tuple(ch_mult)
        self.down = nn.ModuleList()
        for i_level in range(self.num_resolutions):
            block = nn.ModuleList()
            attn = nn.ModuleList()
            block_in = ch*in_ch_mult[i_level]
            block_out = ch*ch_mult[i_level]
            for i_block in range(self.num_res_blocks):
                block.append(ResnetBlock(in_channels=block_in,
                                         out_channels=block_out,
                                         temb_channels=self.temb_ch,
                                         dropout=dropout))
                block_in = block_out
                if curr_res in attn_resolutions:
                    attn.append(AttnBlock(block_in))
            down = nn.Module()
            down.block = block
            down.attn = attn
            if i_level != self.num_resolutions-1:
                down.downsample = Downsample(block_in, resamp_with_conv)
                curr_res = curr_res // 2
            self.down.append(down)

        # middle
        self.mid = nn.Module()
        self.mid.block_1 = ResnetBlock(in_channels=block_in,
                                       out_channels=block_in,
                                       temb_channels=self.temb_ch,
                                       dropout=dropout)
        self.mid.attn_1 = AttnBlock(block_in)
        self.mid.block_2 = ResnetBlock(in_channels=block_in,
                                       out_channels=block_in,
                                       temb_channels=self.temb_ch,
                                       dropout=dropout)

        # end
        self.norm_out = Normalize(block_in)
        self.conv_out = torch.nn.Conv2d(block_in,
                                        2*z_channels if double_z else z_channels,
                                        kernel_size=3,
                                        stride=1,
                                        padding=1)


    def forward(self, x):
        #assert x.shape[2] == x.shape[3] == self.resolution, "{}, {}, {}".format(x.shape[2], x.shape[3], self.resolution)

        # timestep embedding
        temb = None

        # downsampling
        hs = [self.conv_in(x)]
        for i_level in range(self.num_resolutions):
            for i_block in range(self.num_res_blocks):
                h = self.down[i_level].block[i_block](hs[-1], temb)
                if len(self.down[i_level].attn) > 0:
                    h = self.down[i_level].attn[i_block](h)
                hs.append(h)
            if i_level != self.num_resolutions-1:
                hs.append(self.down[i_level].downsample(hs[-1]))

        # middle
        h = hs[-1]
        h = self.mid.block_1(h, temb)
        h = self.mid.attn_1(h)
        h = self.mid.block_2(h, temb)

        # end
        h = self.norm_out(h)
        h = nonlinearity(h)
        h = self.conv_out(h)
        return h


class Decoder(nn.Module):
    def __init__(self, *, ch, out_ch, ch_mult=(1,2,4,8), num_res_blocks,
                 attn_resolutions, dropout=0.0, resamp_with_conv=True, in_channels,
                 resolution, z_channels, give_pre_end=False, **ignorekwargs):
        super().__init__()
        self.ch = ch
        self.temb_ch = 0
        self.num_resolutions = len(ch_mult)
        self.num_res_blocks = num_res_blocks
        self.resolution = resolution
        self.in_channels = in_channels
        self.give_pre_end = give_pre_end

        # compute in_ch_mult, block_in and curr_res at lowest res
        in_ch_mult = (1,)+tuple(ch_mult)
        block_in = ch*ch_mult[self.num_resolutions-1]
        curr_res = resolution // 2**(self.num_resolutions-1)
        self.z_shape = (1,z_channels,curr_res,curr_res)
        print("Working with z of shape {} = {} dimensions.".format(
            self.z_shape, np.prod(self.z_shape)))

        # z to block_in
        self.conv_in = torch.nn.Conv2d(z_channels,
                                       block_in,
                                       kernel_size=3,
                                       stride=1,
                                       padding=1)

        # middle
        self.mid = nn.Module()
        self.mid.block_1 = ResnetBlock(in_channels=block_in,
                                       out_channels=block_in,
                                       temb_channels=self.temb_ch,
                                       dropout=dropout)
        self.mid.attn_1 = AttnBlock(block_in)
        self.mid.block_2 = ResnetBlock(in_channels=block_in,
                                       out_channels=block_in,
                                       temb_channels=self.temb_ch,
                                       dropout=dropout)

        # upsampling
        self.up = nn.ModuleList()
        for i_level in reversed(range(self.num_resolutions)):
            block = nn.ModuleList()
            attn = nn.ModuleList()
            block_out = ch*ch_mult[i_level]
            for i_block in range(self.num_res_blocks+1):
                block.append(ResnetBlock(in_channels=block_in,
                                         out_channels=block_out,
                                         temb_channels=self.temb_ch,
                                         dropout=dropout))
                block_in = block_out
                if curr_res in attn_resolutions:
                    attn.append(AttnBlock(block_in))
            up = nn.Module()
            up.block = block
            up.attn = attn
            if i_level != 0:
                up.upsample = Upsample(block_in, resamp_with_conv)
                curr_res = curr_res * 2
            self.up.insert(0, up) # prepend to get consistent order

        # end
        self.norm_out = Normalize(block_in)
        self.conv_out = torch.nn.Conv2d(block_in,
                                        out_ch,
                                        kernel_size=3,
                                        stride=1,
                                        padding=1)

    def forward(self, z):
        #assert z.shape[1:] == self.z_shape[1:]
        self.last_z_shape = z.shape

        # timestep embedding
        temb = None

        # z to block_in
        h = self.conv_in(z)

        # middle
        h = self.mid.block_1(h, temb)
        h = self.mid.attn_1(h)
        h = self.mid.block_2(h, temb)

        # upsampling
        for i_level in reversed(range(self.num_resolutions)):
            for i_block in range(self.num_res_blocks+1):
                h = self.up[i_level].block[i_block](h, temb)
                if len(self.up[i_level].attn) > 0:
                    h = self.up[i_level].attn[i_block](h)
            if i_level != 0:
                h = self.up[i_level].upsample(h)

        # end
        if self.give_pre_end:
            return h

        h = self.norm_out(h)
        h = nonlinearity(h)
        h = self.conv_out(h)
        return h


class VectorQuantizer(nn.Module):
    """
    Improved version over VectorQuantizer, can be used as a drop-in replacement. Mostly
    avoids costly matrix multiplications and allows for post-hoc remapping of indices.
    """
    # NOTE: due to a bug the beta term was applied to the wrong term. for
    # backwards compatibility we use the buggy version by default, but you can
    # specify legacy=False to fix it.
    def __init__(self, n_e, e_dim, beta, remap=None, unknown_index="random",
                 sane_index_shape=False, legacy=True):
        super().__init__()
        self.n_e = n_e
        self.e_dim = e_dim
        self.beta = beta
        self.legacy = legacy

        self.embedding = nn.Embedding(self.n_e, self.e_dim)
        self.embedding.weight.data.uniform_(-1.0 / self.n_e, 1.0 / self.n_e)

        self.remap = remap
        if self.remap is not None:
            self.used = torch.nn.Parameter(torch.tensor(np.load(self.remap)), requires_grad=False)
            #self.register_buffer("used", torch.tensor(np.load(self.remap)))
            self.re_embed = self.used.shape[0]
            self.unknown_index = unknown_index # "random" or "extra" or integer
            if self.unknown_index == "extra":
                self.unknown_index = self.re_embed
                self.re_embed = self.re_embed+1
            print(f"Remapping {self.n_e} indices to {self.re_embed} indices. "
                  f"Using {self.unknown_index} for unknown indices.")
        else:
            self.re_embed = n_e

        self.sane_index_shape = sane_index_shape

    def remap_to_used(self, inds):
        ishape = inds.shape
        assert len(ishape)>1
        inds = inds.reshape(ishape[0],-1)
        used = self.used.to(inds)
        match = (inds[:,:,None]==used[None,None,...]).long()
        new = match.argmax(-1)
        unknown = match.sum(2)<1
        if self.unknown_index == "random":
            new[unknown]=torch.randint(0,self.re_embed,size=new[unknown].shape).to(device=new.device)
        else:
            new[unknown] = self.unknown_index
        return new.reshape(ishape)

    def unmap_to_all(self, inds):
        ishape = inds.shape
        assert len(ishape)>1
        inds = inds.reshape(ishape[0],-1)
        used = self.used.to(inds)
        if self.re_embed > self.used.shape[0]: # extra token
            inds[inds>=self.used.shape[0]] = 0 # simply set to zero
        back=torch.gather(used[None,:][inds.shape[0]*[0],:], 1, inds)
        return back.reshape(ishape)

    def forward(self, z, temp=None, rescale_logits=False, return_logits=False):
        assert temp is None or temp==1.0, "Only for interface compatible with Gumbel"
        assert rescale_logits==False, "Only for interface compatible with Gumbel"
        assert return_logits==False, "Only for interface compatible with Gumbel"
        # reshape z -> (batch, height, width, channel) and flatten
        z = rearrange(z, 'b c h w -> b h w c').contiguous()
        z_flattened = z.view(-1, self.e_dim)
        # distances from z to embeddings e_j (z - e)^2 = z^2 + e^2 - 2 e * z

        d = torch.sum(z_flattened ** 2, dim=1, keepdim=True) + \
            torch.sum(self.embedding.weight**2, dim=1) - 2 * \
            torch.einsum('bd,dn->bn', z_flattened, rearrange(self.embedding.weight, 'n d -> d n'))

        min_encoding_indices = torch.argmin(d, dim=1)
        z_q = self.embedding(min_encoding_indices).view(z.shape)
        perplexity = None
        min_encodings = None

        # compute loss for embedding
        if not self.legacy:
            loss = self.beta * torch.mean((z_q.detach()-z)**2) + \
                   torch.mean((z_q - z.detach()) ** 2)
        else:
            loss = torch.mean((z_q.detach()-z)**2) + self.beta * \
                   torch.mean((z_q - z.detach()) ** 2)

        # preserve gradients
        z_q = z + (z_q - z).detach()

        # reshape back to match original input shape
        z_q = rearrange(z_q, 'b h w c -> b c h w').contiguous()

        if self.remap is not None:
            min_encoding_indices = min_encoding_indices.reshape(z.shape[0],-1) # add batch axis
            min_encoding_indices = self.remap_to_used(min_encoding_indices)
            min_encoding_indices = min_encoding_indices.reshape(-1,1) # flatten

        if self.sane_index_shape:
            min_encoding_indices = min_encoding_indices.reshape(
                z_q.shape[0], z_q.shape[2], z_q.shape[3])

        return z_q, loss, (perplexity, min_encodings, min_encoding_indices)

    def get_codebook_entry(self, indices, shape):
        # shape specifying (batch, height, width, channel)
        if self.remap is not None:
            indices = indices.reshape(shape[0],-1) # add batch axis
            indices = self.unmap_to_all(indices)
            indices = indices.reshape(-1) # flatten again

        # get quantized latent vectors
        z_q = self.embedding(indices)

        if shape is not None:
            z_q = z_q.view(shape)
            # reshape back to match original input shape
            z_q = z_q.permute(0, 3, 1, 2).contiguous()

        return z_q


def get_obj_from_str(string, reload=False):
    module, cls = string.rsplit(".", 1)
    if reload:
        module_imp = importlib.import_module(module)
        importlib.reload(module_imp)
    return getattr(importlib.import_module(module, package=None), cls)


def instantiate_from_config(config):
    if not "target" in config:
        raise KeyError("Expected key `target` to instantiate.")
    return get_obj_from_str(config["target"])(**config.get("params", dict()))


class VQModelPytorch(torch.nn.Module):
    def __init__(self,
                 ddconfig,
                 n_embed,
                 embed_dim,
                 ckpt_path=None,
                 ignore_keys=[],
                 image_key="image",
                 colorize_nlabels=None,
                 monitor=None,
                 remap=None,
                 sane_index_shape=False,  # tell vector quantizer to return indices as bhw
                 ):
        super().__init__()
        self.image_key = image_key
        self.encoder = Encoder(**ddconfig)
        self.decoder = Decoder(**ddconfig)
        self.quantize = VectorQuantizer(n_embed, embed_dim, beta=0.25,
                                        remap=remap, sane_index_shape=sane_index_shape)
        self.quant_conv = torch.nn.Conv2d(ddconfig["z_channels"], embed_dim, 1)
        self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1)

        if ckpt_path is not None:
            self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys)
        self.image_key = image_key
        if colorize_nlabels is not None:
            assert type(colorize_nlabels)==int
            self.register_buffer("colorize", torch.randn(3, colorize_nlabels, 1, 1))
        if monitor is not None:
            self.monitor = monitor     

    def init_from_ckpt(self, path, ignore_keys=list()):
        sd = torch.load(path, map_location="cpu")["state_dict"]
        keys = list(sd.keys())
        for k in keys:
            for ik in ignore_keys:
                if k.startswith(ik):
                    print("Deleting key {} from state_dict.".format(k))
                    del sd[k]
        self.load_state_dict(sd, strict=False)
        print(f"Restored from {path}")

    def encode(self, x):
        h = self.encoder(x)
        h = self.quant_conv(h)
        quant, emb_loss, info = self.quantize(h)
        return quant, emb_loss, info

    def decode(self, quant):
        quant = self.post_quant_conv(quant)
        dec = self.decoder(quant)
        return dec

    def decode_code(self, code_b):
        quant_b = self.quantize.embed_code(code_b)
        dec = self.decode(quant_b)
        return dec

    def forward(self, input):
        #print("Encoding..")
        quant, diff, _ = self.encode(input)
        #print("Decoding..")
        dec = self.decode(quant)
        #print("Returning..")
        return dec, diff

    def to_rgb(self, x):
        assert self.image_key == "segmentation"
        if not hasattr(self, "colorize"):
            print('Creating colorize in to_rgb()')
            self.colorize = torch.nn.Parameter(torch.randn(3, x.shape[1], 1, 1).to(x.clone()), requires_grad=False)
            #self.register_buffer("colorize", torch.randn(3, x.shape[1], 1, 1).to(x.clone()))
            print('Created colorize in to_rgb()')
        x = F.conv2d(x, weight=self.colorize)
        x = 2.*(x.clone()-x.min())/(x.max()-x.min()) - 1.
        return x

    def get_input(self, batch, k):
        x = batch[k].clone()
        if len(x.shape) == 3:
            x = x[..., None].clone()
        x = x.clone().permute(0, 3, 1, 2).to(memory_format=torch.contiguous_format)
        return x.float()
def batch(image, processor):
    if not image.mode == "RGB":
        image = image.convert("RGB")
    image = np.array(image).astype(np.uint8)
    processed = processor(image=image)
    return {
        "image": (processed["image"]/127.5 - 1.0).astype(np.float32)
    
    }
def sample(model, batch):
    x = model.get_input(batch, 'image')
    xrec, _ = model(x)
    if x.shape[1] > 3:
      assert xrec.shape[1] > 3
      x = model.to_rgb(x)
      xrec = model.to_rgb(xrec)
    return x, xrec

def save_image(x, path):
     c,h,w = x.shape
     assert c==3
     x = ((x.detach().cpu().numpy().transpose(1,2,0)+1.0)*127.5).clip(0,255).astype(np.uint8)
     Image.fromarray(x).save(path)

if __name__ == "__main__":
    import sys

    print(sys.argv)
    ckpt_path = sys.argv[1]
    ddconfig = {
      "double_z": False,
      "z_channels": 256,
      "resolution": 336,
      "in_channels": 3,
      "out_ch": 3,
      "ch": 128,
      "ch_mult": [ 1,1,2,2,4],  # num_down = len(ch_mult)-1
      "num_res_blocks": 2,
      "attn_resolutions": [21],
      "dropout": 0.0,
    }

    model = VQModelPytorch(
        ddconfig=ddconfig,
        n_embed=16384,
        embed_dim=256,
        ckpt_path=ckpt_path,
        ignore_keys=['loss']
    )
    from PIL import Image
    image = Image.new('RGB', (400, 400))
    print(image.size)
    rescaler = albumentations.SmallestMaxSize(max_size=336)
    cropper = albumentations.RandomCrop(
      height=336, 
      width=336)
    preprocessor = albumentations.Compose(
      [rescaler, cropper])
    inp = batch(image, processor=preprocessor)
    x, xrec = sample(model, custom_collate([inp]))
    save_image(xrec[0], 'tmp.png')
    print(model)
    import time
    start = time.time()
    print(model.encode(x))
    end = time.time()
    print("took: ", end-start)