|
from abc import abstractmethod |
|
from typing import Iterable |
|
|
|
import numpy as np |
|
import torch as th |
|
import torch.nn as nn |
|
import torch.nn.functional as F |
|
from einops import rearrange |
|
|
|
from ...modules.attention import SpatialTransformer |
|
from ...modules.diffusionmodules.util import ( |
|
avg_pool_nd, |
|
conv_nd, |
|
linear, |
|
normalization, |
|
timestep_embedding, |
|
zero_module, |
|
) |
|
from ...util import default, exists |
|
|
|
|
|
class Timestep(nn.Module): |
|
def __init__(self, dim): |
|
super().__init__() |
|
self.dim = dim |
|
|
|
def forward(self, t): |
|
return timestep_embedding(t, self.dim) |
|
|
|
|
|
class TimestepBlock(nn.Module): |
|
""" |
|
Any module where forward() takes timestep embeddings as a second argument. |
|
""" |
|
|
|
@abstractmethod |
|
def forward(self, x, emb): |
|
""" |
|
Apply the module to `x` given `emb` timestep embeddings. |
|
""" |
|
|
|
|
|
class TimestepEmbedSequential(nn.Sequential, TimestepBlock): |
|
""" |
|
A sequential module that passes timestep embeddings to the children that |
|
support it as an extra input. |
|
""" |
|
|
|
def forward( |
|
self, |
|
x, |
|
emb, |
|
t_context=None, |
|
v_context=None |
|
): |
|
for layer in self: |
|
if isinstance(layer, TimestepBlock): |
|
x = layer(x, emb) |
|
elif isinstance(layer, SpatialTransformer): |
|
x = layer(x, t_context, v_context) |
|
else: |
|
x = layer(x) |
|
return x |
|
|
|
|
|
class Upsample(nn.Module): |
|
""" |
|
An upsampling layer with an optional convolution. |
|
:param channels: channels in the inputs and outputs. |
|
:param use_conv: a bool determining if a convolution is applied. |
|
:param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then |
|
upsampling occurs in the inner-two dimensions. |
|
""" |
|
|
|
def __init__( |
|
self, channels, use_conv, dims=2, out_channels=None, padding=1, third_up=False |
|
): |
|
super().__init__() |
|
self.channels = channels |
|
self.out_channels = out_channels or channels |
|
self.use_conv = use_conv |
|
self.dims = dims |
|
self.third_up = third_up |
|
if use_conv: |
|
self.conv = conv_nd( |
|
dims, self.channels, self.out_channels, 3, padding=padding |
|
) |
|
|
|
def forward(self, x): |
|
assert x.shape[1] == self.channels |
|
if self.dims == 3: |
|
t_factor = 1 if not self.third_up else 2 |
|
x = F.interpolate( |
|
x, |
|
(t_factor * x.shape[2], x.shape[3] * 2, x.shape[4] * 2), |
|
mode="nearest", |
|
) |
|
else: |
|
x = F.interpolate(x, scale_factor=2, mode="nearest") |
|
if self.use_conv: |
|
x = self.conv(x) |
|
return x |
|
|
|
|
|
class Downsample(nn.Module): |
|
""" |
|
A downsampling layer with an optional convolution. |
|
:param channels: channels in the inputs and outputs. |
|
:param use_conv: a bool determining if a convolution is applied. |
|
:param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then |
|
downsampling occurs in the inner-two dimensions. |
|
""" |
|
|
|
def __init__( |
|
self, channels, use_conv, dims=2, out_channels=None, padding=1, third_down=False |
|
): |
|
super().__init__() |
|
self.channels = channels |
|
self.out_channels = out_channels or channels |
|
self.use_conv = use_conv |
|
self.dims = dims |
|
stride = 2 if dims != 3 else ((1, 2, 2) if not third_down else (2, 2, 2)) |
|
if use_conv: |
|
|
|
|
|
|
|
|
|
|
|
if dims == 3: |
|
pass |
|
|
|
self.op = conv_nd( |
|
dims, |
|
self.channels, |
|
self.out_channels, |
|
3, |
|
stride=stride, |
|
padding=padding, |
|
) |
|
else: |
|
assert self.channels == self.out_channels |
|
self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride) |
|
|
|
def forward(self, x): |
|
assert x.shape[1] == self.channels |
|
return self.op(x) |
|
|
|
|
|
class ResBlock(TimestepBlock): |
|
""" |
|
A residual block that can optionally change the number of channels. |
|
""" |
|
|
|
def __init__( |
|
self, |
|
channels, |
|
emb_channels, |
|
dropout, |
|
out_channels=None, |
|
use_conv=False, |
|
use_scale_shift_norm=False, |
|
dims=2, |
|
up=False, |
|
down=False, |
|
kernel_size=3, |
|
exchange_temb_dims=False, |
|
skip_t_emb=False |
|
): |
|
super().__init__() |
|
self.channels = channels |
|
self.emb_channels = emb_channels |
|
self.dropout = dropout |
|
self.out_channels = out_channels or channels |
|
self.use_conv = use_conv |
|
self.use_scale_shift_norm = use_scale_shift_norm |
|
self.exchange_temb_dims = exchange_temb_dims |
|
|
|
if isinstance(kernel_size, Iterable): |
|
padding = [k // 2 for k in kernel_size] |
|
else: |
|
padding = kernel_size // 2 |
|
|
|
self.in_layers = nn.Sequential( |
|
normalization(channels), |
|
nn.SiLU(), |
|
conv_nd(dims, channels, self.out_channels, kernel_size, padding=padding), |
|
) |
|
|
|
self.updown = up or down |
|
|
|
if up: |
|
self.h_upd = Upsample(channels, False, dims) |
|
self.x_upd = Upsample(channels, False, dims) |
|
elif down: |
|
self.h_upd = Downsample(channels, False, dims) |
|
self.x_upd = Downsample(channels, False, dims) |
|
else: |
|
self.h_upd = self.x_upd = nn.Identity() |
|
|
|
self.skip_t_emb = skip_t_emb |
|
self.emb_out_channels = ( |
|
2 * self.out_channels if use_scale_shift_norm else self.out_channels |
|
) |
|
if self.skip_t_emb: |
|
print(f"Skipping timestep embedding in {self.__class__.__name__}") |
|
assert not self.use_scale_shift_norm |
|
self.emb_layers = None |
|
self.exchange_temb_dims = False |
|
else: |
|
self.emb_layers = nn.Sequential( |
|
nn.SiLU(), |
|
linear( |
|
emb_channels, |
|
self.emb_out_channels, |
|
), |
|
) |
|
|
|
self.out_layers = nn.Sequential( |
|
normalization(self.out_channels), |
|
nn.SiLU(), |
|
nn.Dropout(p=dropout), |
|
zero_module( |
|
conv_nd( |
|
dims, |
|
self.out_channels, |
|
self.out_channels, |
|
kernel_size, |
|
padding=padding, |
|
) |
|
), |
|
) |
|
|
|
if self.out_channels == channels: |
|
self.skip_connection = nn.Identity() |
|
elif use_conv: |
|
self.skip_connection = conv_nd( |
|
dims, channels, self.out_channels, kernel_size, padding=padding |
|
) |
|
else: |
|
self.skip_connection = conv_nd(dims, channels, self.out_channels, 1) |
|
|
|
def forward(self, x, emb): |
|
if self.updown: |
|
in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1] |
|
h = in_rest(x) |
|
h = self.h_upd(h) |
|
x = self.x_upd(x) |
|
h = in_conv(h) |
|
else: |
|
h = self.in_layers(x) |
|
|
|
if self.skip_t_emb: |
|
emb_out = th.zeros_like(h) |
|
else: |
|
emb_out = self.emb_layers(emb).type(h.dtype) |
|
while len(emb_out.shape) < len(h.shape): |
|
emb_out = emb_out[..., None] |
|
if self.use_scale_shift_norm: |
|
out_norm, out_rest = self.out_layers[0], self.out_layers[1:] |
|
scale, shift = th.chunk(emb_out, 2, dim=1) |
|
h = out_norm(h) * (1 + scale) + shift |
|
h = out_rest(h) |
|
else: |
|
if self.exchange_temb_dims: |
|
emb_out = rearrange(emb_out, "b t c ... -> b c t ...") |
|
h = h + emb_out |
|
h = self.out_layers(h) |
|
return self.skip_connection(x) + h |
|
|
|
|
|
import seaborn as sns |
|
import matplotlib.pyplot as plt |
|
|
|
|
|
class UnifiedUNetModel(nn.Module): |
|
|
|
def __init__( |
|
self, |
|
in_channels, |
|
ctrl_channels, |
|
model_channels, |
|
out_channels, |
|
num_res_blocks, |
|
attention_resolutions, |
|
dropout=0, |
|
channel_mult=(1, 2, 4, 8), |
|
save_attn_type=None, |
|
save_attn_layers=[], |
|
conv_resample=True, |
|
dims=2, |
|
use_label=None, |
|
num_heads=-1, |
|
num_head_channels=-1, |
|
num_heads_upsample=-1, |
|
use_scale_shift_norm=False, |
|
resblock_updown=False, |
|
transformer_depth=1, |
|
t_context_dim=None, |
|
v_context_dim=None, |
|
num_attention_blocks=None, |
|
use_linear_in_transformer=False, |
|
adm_in_channels=None, |
|
transformer_depth_middle=None |
|
): |
|
super().__init__() |
|
|
|
if num_heads_upsample == -1: |
|
num_heads_upsample = num_heads |
|
|
|
if num_heads == -1: |
|
assert ( |
|
num_head_channels != -1 |
|
), "Either num_heads or num_head_channels has to be set" |
|
|
|
if num_head_channels == -1: |
|
assert ( |
|
num_heads != -1 |
|
), "Either num_heads or num_head_channels has to be set" |
|
|
|
self.in_channels = in_channels |
|
self.ctrl_channels = ctrl_channels |
|
self.model_channels = model_channels |
|
self.out_channels = out_channels |
|
|
|
transformer_depth = len(channel_mult) * [transformer_depth] |
|
transformer_depth_middle = default(transformer_depth_middle, transformer_depth[-1]) |
|
|
|
self.num_res_blocks = len(channel_mult) * [num_res_blocks] |
|
|
|
self.attention_resolutions = attention_resolutions |
|
self.dropout = dropout |
|
self.channel_mult = channel_mult |
|
self.conv_resample = conv_resample |
|
self.use_label = use_label |
|
self.num_heads = num_heads |
|
self.num_head_channels = num_head_channels |
|
self.num_heads_upsample = num_heads_upsample |
|
|
|
time_embed_dim = model_channels * 4 |
|
self.time_embed = nn.Sequential( |
|
linear(model_channels, time_embed_dim), |
|
nn.SiLU(), |
|
linear(time_embed_dim, time_embed_dim), |
|
) |
|
|
|
if self.use_label is not None: |
|
self.label_emb = nn.Sequential( |
|
nn.Sequential( |
|
linear(adm_in_channels, time_embed_dim), |
|
nn.SiLU(), |
|
linear(time_embed_dim, time_embed_dim), |
|
) |
|
) |
|
|
|
self.input_blocks = nn.ModuleList( |
|
[ |
|
TimestepEmbedSequential( |
|
conv_nd(dims, in_channels, model_channels, 3, padding=1) |
|
) |
|
] |
|
) |
|
|
|
if self.ctrl_channels > 0: |
|
self.ctrl_block = TimestepEmbedSequential( |
|
conv_nd(dims, ctrl_channels, 16, 3, padding=1), |
|
nn.SiLU(), |
|
conv_nd(dims, 16, 16, 3, padding=1), |
|
nn.SiLU(), |
|
conv_nd(dims, 16, 32, 3, padding=1), |
|
nn.SiLU(), |
|
conv_nd(dims, 32, 32, 3, padding=1), |
|
nn.SiLU(), |
|
conv_nd(dims, 32, 96, 3, padding=1), |
|
nn.SiLU(), |
|
conv_nd(dims, 96, 96, 3, padding=1), |
|
nn.SiLU(), |
|
conv_nd(dims, 96, 256, 3, padding=1), |
|
nn.SiLU(), |
|
zero_module(conv_nd(dims, 256, model_channels, 3, padding=1)) |
|
) |
|
|
|
self._feature_size = model_channels |
|
input_block_chans = [model_channels] |
|
ch = model_channels |
|
ds = 1 |
|
for level, mult in enumerate(channel_mult): |
|
for nr in range(self.num_res_blocks[level]): |
|
layers = [ |
|
ResBlock( |
|
ch, |
|
time_embed_dim, |
|
dropout, |
|
out_channels=mult * model_channels, |
|
dims=dims, |
|
use_scale_shift_norm=use_scale_shift_norm |
|
) |
|
] |
|
ch = mult * model_channels |
|
if ds in attention_resolutions: |
|
if num_head_channels == -1: |
|
dim_head = ch // num_heads |
|
else: |
|
num_heads = ch // num_head_channels |
|
dim_head = num_head_channels |
|
if ( |
|
not exists(num_attention_blocks) |
|
or nr < num_attention_blocks[level] |
|
): |
|
layers.append( |
|
SpatialTransformer( |
|
ch, |
|
num_heads, |
|
dim_head, |
|
depth=transformer_depth[level], |
|
t_context_dim=t_context_dim, |
|
v_context_dim=v_context_dim, |
|
use_linear=use_linear_in_transformer |
|
) |
|
) |
|
self.input_blocks.append(TimestepEmbedSequential(*layers)) |
|
self._feature_size += ch |
|
input_block_chans.append(ch) |
|
if level != len(channel_mult) - 1: |
|
out_ch = ch |
|
self.input_blocks.append( |
|
TimestepEmbedSequential( |
|
ResBlock( |
|
ch, |
|
time_embed_dim, |
|
dropout, |
|
out_channels=out_ch, |
|
dims=dims, |
|
use_scale_shift_norm=use_scale_shift_norm, |
|
down=True |
|
) |
|
if resblock_updown |
|
else Downsample( |
|
ch, conv_resample, dims=dims, out_channels=out_ch |
|
) |
|
) |
|
) |
|
ch = out_ch |
|
input_block_chans.append(ch) |
|
ds *= 2 |
|
self._feature_size += ch |
|
|
|
if num_head_channels == -1: |
|
dim_head = ch // num_heads |
|
else: |
|
num_heads = ch // num_head_channels |
|
dim_head = num_head_channels |
|
|
|
self.middle_block = TimestepEmbedSequential( |
|
ResBlock( |
|
ch, |
|
time_embed_dim, |
|
dropout, |
|
dims=dims, |
|
use_scale_shift_norm=use_scale_shift_norm |
|
), |
|
SpatialTransformer( |
|
ch, |
|
num_heads, |
|
dim_head, |
|
depth=transformer_depth_middle, |
|
t_context_dim=t_context_dim, |
|
v_context_dim=v_context_dim, |
|
use_linear=use_linear_in_transformer |
|
), |
|
ResBlock( |
|
ch, |
|
time_embed_dim, |
|
dropout, |
|
dims=dims, |
|
use_scale_shift_norm=use_scale_shift_norm |
|
) |
|
) |
|
|
|
self._feature_size += ch |
|
|
|
self.output_blocks = nn.ModuleList([]) |
|
for level, mult in list(enumerate(channel_mult))[::-1]: |
|
for i in range(self.num_res_blocks[level] + 1): |
|
ich = input_block_chans.pop() |
|
layers = [ |
|
ResBlock( |
|
ch + ich, |
|
time_embed_dim, |
|
dropout, |
|
out_channels=model_channels * mult, |
|
dims=dims, |
|
use_scale_shift_norm=use_scale_shift_norm |
|
) |
|
] |
|
ch = model_channels * mult |
|
if ds in attention_resolutions: |
|
if num_head_channels == -1: |
|
dim_head = ch // num_heads |
|
else: |
|
num_heads = ch // num_head_channels |
|
dim_head = num_head_channels |
|
if ( |
|
not exists(num_attention_blocks) |
|
or i < num_attention_blocks[level] |
|
): |
|
layers.append( |
|
SpatialTransformer( |
|
ch, |
|
num_heads, |
|
dim_head, |
|
depth=transformer_depth[level], |
|
t_context_dim=t_context_dim, |
|
v_context_dim=v_context_dim, |
|
use_linear=use_linear_in_transformer |
|
) |
|
) |
|
if level and i == self.num_res_blocks[level]: |
|
out_ch = ch |
|
layers.append( |
|
ResBlock( |
|
ch, |
|
time_embed_dim, |
|
dropout, |
|
out_channels=out_ch, |
|
dims=dims, |
|
use_scale_shift_norm=use_scale_shift_norm, |
|
up=True |
|
) |
|
if resblock_updown |
|
else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch) |
|
) |
|
ds //= 2 |
|
self.output_blocks.append(TimestepEmbedSequential(*layers)) |
|
self._feature_size += ch |
|
|
|
self.out = nn.Sequential( |
|
normalization(ch), |
|
nn.SiLU(), |
|
zero_module(conv_nd(dims, model_channels, out_channels, 3, padding=1)) |
|
) |
|
|
|
|
|
self.attn_type = save_attn_type |
|
self.attn_layers = save_attn_layers |
|
self.attn_map_cache = [] |
|
for name, module in self.named_modules(): |
|
if any([name.endswith(attn_type) for attn_type in self.attn_type]): |
|
item = {"name": name, "heads": module.heads, "size": None, "attn_map": None} |
|
self.attn_map_cache.append(item) |
|
module.attn_map_cache = item |
|
|
|
def clear_attn_map(self): |
|
|
|
for item in self.attn_map_cache: |
|
if item["attn_map"] is not None: |
|
del item["attn_map"] |
|
item["attn_map"] = None |
|
|
|
def save_attn_map(self, attn_type="t_attn", save_name="temp", tokens=""): |
|
|
|
attn_maps = [] |
|
for item in self.attn_map_cache: |
|
name = item["name"] |
|
if any([name.startswith(block) for block in self.attn_layers]) and name.endswith(attn_type): |
|
heads = item["heads"] |
|
attn_maps.append(item["attn_map"].detach().cpu()) |
|
|
|
attn_map = th.stack(attn_maps, dim=0) |
|
attn_map = th.mean(attn_map, dim=0) |
|
|
|
|
|
bh, n, l = attn_map.shape |
|
attn_map = attn_map.reshape((-1,heads,n,l)).mean(dim=1) |
|
b = attn_map.shape[0] |
|
|
|
h = w = int(n**0.5) |
|
attn_map = attn_map.permute(0,2,1).reshape((b,l,h,w)).numpy() |
|
attn_map_i = attn_map[-1] |
|
|
|
l = attn_map_i.shape[0] |
|
fig = plt.figure(figsize=(12, 8), dpi=300) |
|
for j in range(12): |
|
if j >= l: break |
|
ax = fig.add_subplot(3, 4, j+1) |
|
sns.heatmap(attn_map_i[j], square=True, xticklabels=False, yticklabels=False) |
|
if j < len(tokens): |
|
ax.set_title(tokens[j]) |
|
fig.savefig(f"temp/attn_map/attn_map_{save_name}.png") |
|
plt.close() |
|
|
|
return attn_map_i |
|
|
|
def forward(self, x, timesteps=None, t_context=None, v_context=None, y=None, **kwargs): |
|
|
|
assert (y is not None) == ( |
|
self.use_label is not None |
|
), "must specify y if and only if the model is class-conditional" |
|
|
|
self.clear_attn_map() |
|
|
|
hs = [] |
|
t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False) |
|
emb = self.time_embed(t_emb) |
|
|
|
if self.use_label is not None: |
|
assert y.shape[0] == x.shape[0] |
|
emb = emb + self.label_emb(y) |
|
|
|
h = x |
|
if self.ctrl_channels > 0: |
|
in_h, add_h = th.split(h, [self.in_channels, self.ctrl_channels], dim=1) |
|
for i, module in enumerate(self.input_blocks): |
|
if self.ctrl_channels > 0 and i == 0: |
|
h = module(in_h, emb, t_context, v_context) + self.ctrl_block(add_h, emb, t_context, v_context) |
|
else: |
|
h = module(h, emb, t_context, v_context) |
|
hs.append(h) |
|
h = self.middle_block(h, emb, t_context, v_context) |
|
for i, module in enumerate(self.output_blocks): |
|
h = th.cat([h, hs.pop()], dim=1) |
|
h = module(h, emb, t_context, v_context) |
|
h = h.type(x.dtype) |
|
|
|
return self.out(h) |