DL-Art-School/codes/models/archs/SwitchedResidualGenerator_arch.py
2020-09-15 16:55:38 -06:00

521 lines
26 KiB
Python

import torch
from torch import nn
from switched_conv import BareConvSwitch, compute_attention_specificity, AttentionNorm
import torch.nn.functional as F
import functools
from collections import OrderedDict
from models.archs.arch_util import ConvBnLelu, ConvGnSilu, ExpansionBlock, ExpansionBlock2, ConvGnLelu, MultiConvBlock
from switched_conv_util import save_attention_to_image_rgb
import os
from torch.utils.checkpoint import checkpoint
from models.archs.spinenet_arch import SpineNet
# Set to true to relieve memory pressure by using torch.utils.checkpoint in several memory-critical locations.
memory_checkpointing_enabled = True
# VGG-style layer with Conv(stride2)->BN->Activation->Conv->BN->Activation
# Doubles the input filter count.
class HalvingProcessingBlock(nn.Module):
def __init__(self, filters):
super(HalvingProcessingBlock, self).__init__()
self.bnconv1 = ConvGnSilu(filters, filters * 2, stride=2, norm=False, bias=False)
self.bnconv2 = ConvGnSilu(filters * 2, filters * 2, norm=True, bias=False)
def forward(self, x):
x = self.bnconv1(x)
return self.bnconv2(x)
# This is a classic u-net architecture with the goal of assigning each individual pixel an individual transform
# switching set.
class ConvBasisMultiplexer(nn.Module):
def __init__(self, input_channels, base_filters, reductions, processing_depth, multiplexer_channels, use_gn=True, use_exp2=False):
super(ConvBasisMultiplexer, self).__init__()
self.filter_conv = ConvGnSilu(input_channels, base_filters, bias=True)
self.reduction_blocks = nn.ModuleList([HalvingProcessingBlock(base_filters * 2 ** i) for i in range(reductions)])
reduction_filters = base_filters * 2 ** reductions
self.processing_blocks = nn.Sequential(OrderedDict([('block%i' % (i,), ConvGnSilu(reduction_filters, reduction_filters, bias=False)) for i in range(processing_depth)]))
if use_exp2:
self.expansion_blocks = nn.ModuleList([ExpansionBlock2(reduction_filters // (2 ** i)) for i in range(reductions)])
else:
self.expansion_blocks = nn.ModuleList([ExpansionBlock(reduction_filters // (2 ** i)) for i in range(reductions)])
gap = base_filters - multiplexer_channels
cbl1_out = ((base_filters - (gap // 2)) // 4) * 4 # Must be multiples of 4 to use with group norm.
self.cbl1 = ConvGnSilu(base_filters, cbl1_out, norm=use_gn, bias=False, num_groups=4)
cbl2_out = ((base_filters - (3 * gap // 4)) // 4) * 4
self.cbl2 = ConvGnSilu(cbl1_out, cbl2_out, norm=use_gn, bias=False, num_groups=4)
self.cbl3 = ConvGnSilu(cbl2_out, multiplexer_channels, bias=True, norm=False)
def forward(self, x):
x = self.filter_conv(x)
reduction_identities = []
for b in self.reduction_blocks:
reduction_identities.append(x)
x = b(x)
x = self.processing_blocks(x)
for i, b in enumerate(self.expansion_blocks):
x = b(x, reduction_identities[-i - 1])
x = self.cbl1(x)
x = self.cbl2(x)
x = self.cbl3(x)
return x
# torch.gather() which operates across 2d images.
def gather_2d(input, index):
b, c, h, w = input.shape
nodim = input.view(b, c, h * w)
ind_nd = index[:, 0]*w + index[:, 1]
ind_nd = ind_nd.unsqueeze(1)
ind_nd = ind_nd.repeat((1, c))
ind_nd = ind_nd.unsqueeze(2)
result = torch.gather(nodim, dim=2, index=ind_nd)
result = result.squeeze()
if b == 1:
result = result.unsqueeze(0)
return result
# Computes a linear latent by performing processing on the reference image and returning the filters of a single point,
# which should be centered on the image patch being processed.
#
# Output is base_filters * 8.
class ReferenceImageBranch(nn.Module):
def __init__(self, base_filters=64):
super(ReferenceImageBranch, self).__init__()
self.filter_conv = ConvGnSilu(4, base_filters, bias=True)
self.reduction_blocks = nn.ModuleList([HalvingProcessingBlock(base_filters * 2 ** i) for i in range(3)])
reduction_filters = base_filters * 2 ** 3
self.processing_blocks = nn.Sequential(OrderedDict([('block%i' % (i,), ConvGnSilu(reduction_filters, reduction_filters, bias=False)) for i in range(4)]))
# center_point is a [b,2] long tensor describing the center point of where the patch was taken from the reference
# image.
def forward(self, x, center_point):
x = self.filter_conv(x)
reduction_identities = []
for b in self.reduction_blocks:
reduction_identities.append(x)
x = b(x)
x = self.processing_blocks(x)
return gather_2d(x, center_point // 8)
class AdaInConvBlock(nn.Module):
def __init__(self, reference_size, in_nc, out_nc, conv_block=ConvGnLelu):
super(AdaInConvBlock, self).__init__()
self.filter_conv = conv_block(in_nc, out_nc, activation=True, norm=False, bias=False)
self.ref_proc = nn.Linear(reference_size, reference_size)
self.ref_red = nn.Linear(reference_size, out_nc * 2)
self.feature_norm = torch.nn.InstanceNorm2d(out_nc)
self.style_norm = torch.nn.InstanceNorm1d(out_nc)
self.post_fuse_conv = conv_block(out_nc, out_nc, activation=False, norm=True, bias=True)
def forward(self, x, ref):
x = self.feature_norm(self.filter_conv(x))
ref = self.ref_proc(ref)
ref = self.ref_red(ref)
b, c = ref.shape
ref = self.style_norm(ref.view(b, 2, c // 2))
x = x * ref[:, 0, :].unsqueeze(dim=2).unsqueeze(dim=3).expand(x.shape) + ref[:, 1, :].unsqueeze(dim=2).unsqueeze(dim=3).expand(x.shape)
return self.post_fuse_conv(x)
class ProcessingBranchWithStochasticity(nn.Module):
def __init__(self, nf_in, nf_out, noise_filters, depth):
super(ProcessingBranchWithStochasticity, self).__init__()
nf_gap = nf_out - nf_in
self.noise_filters = noise_filters
self.processor = MultiConvBlock(nf_in + noise_filters, nf_in + nf_gap // 2, nf_out, kernel_size=3, depth=depth, weight_init_factor = .1)
def forward(self, x):
b, c, h, w = x.shape
noise = torch.randn((b, self.noise_filters, h, w), device=x.device)
return self.processor(torch.cat([x, noise], dim=1))
# This is similar to ConvBasisMultiplexer, except that it takes a linear reference tensor as a second input to
# provide better results. It also has fixed parameterization in several places
class ReferencingConvMultiplexer(nn.Module):
def __init__(self, input_channels, base_filters, multiplexer_channels, use_gn=True):
super(ReferencingConvMultiplexer, self).__init__()
self.style_fuse = AdaInConvBlock(512, input_channels, base_filters, ConvGnSilu)
self.reduction_blocks = nn.ModuleList([HalvingProcessingBlock(base_filters * 2 ** i) for i in range(3)])
reduction_filters = base_filters * 2 ** 3
self.processing_blocks = nn.Sequential(OrderedDict([('block%i' % (i,), ConvGnSilu(reduction_filters, reduction_filters, bias=False)) for i in range(2)]))
self.expansion_blocks = nn.ModuleList([ExpansionBlock2(reduction_filters // (2 ** i)) for i in range(3)])
gap = base_filters - multiplexer_channels
cbl1_out = ((base_filters - (gap // 2)) // 4) * 4 # Must be multiples of 4 to use with group norm.
self.cbl1 = ConvGnSilu(base_filters, cbl1_out, norm=use_gn, bias=False, num_groups=4)
cbl2_out = ((base_filters - (3 * gap // 4)) // 4) * 4
self.cbl2 = ConvGnSilu(cbl1_out, cbl2_out, norm=use_gn, bias=False, num_groups=4)
self.cbl3 = ConvGnSilu(cbl2_out, multiplexer_channels, bias=True, norm=False)
def forward(self, x, ref):
x = self.style_fuse(x, ref)
reduction_identities = []
for b in self.reduction_blocks:
reduction_identities.append(x)
x = b(x)
x = self.processing_blocks(x)
for i, b in enumerate(self.expansion_blocks):
x = b(x, reduction_identities[-i - 1])
x = self.cbl1(x)
x = self.cbl2(x)
x = self.cbl3(x)
return x
class ConfigurableSwitchComputer(nn.Module):
def __init__(self, base_filters, multiplexer_net, pre_transform_block, transform_block, transform_count, attention_norm,
init_temp=20, add_scalable_noise_to_transforms=False, feed_transforms_into_multiplexer=False):
super(ConfigurableSwitchComputer, self).__init__()
tc = transform_count
self.multiplexer = multiplexer_net(tc)
if pre_transform_block:
self.pre_transform = pre_transform_block()
else:
self.pre_transform = None
self.transforms = nn.ModuleList([transform_block() for _ in range(transform_count)])
self.add_noise = add_scalable_noise_to_transforms
self.feed_transforms_into_multiplexer = feed_transforms_into_multiplexer
self.noise_scale = nn.Parameter(torch.full((1,), float(1e-3)))
# And the switch itself, including learned scalars
self.switch = BareConvSwitch(initial_temperature=init_temp, attention_norm=AttentionNorm(transform_count, accumulator_size=16 * transform_count) if attention_norm else None)
self.switch_scale = nn.Parameter(torch.full((1,), float(1)))
self.post_switch_conv = ConvBnLelu(base_filters, base_filters, norm=False, bias=True)
# The post_switch_conv gets a low scale initially. The network can decide to magnify it (or not)
# depending on its needs.
self.psc_scale = nn.Parameter(torch.full((1,), float(.1)))
# Regarding inputs: it is acceptable to pass in a tuple/list as an input for (x), but the first element
# *must* be the actual parameter that gets fed through the network - it is assumed to be the identity.
def forward(self, x, output_attention_weights=False, identity=None, att_in=None, fixed_scale=1):
if isinstance(x, tuple):
x1 = x[0]
else:
x1 = x
if att_in is None:
att_in = x
if identity is None:
identity = x1
if self.add_noise:
rand_feature = torch.randn_like(x1) * self.noise_scale
if isinstance(x, tuple):
x = (x1 + rand_feature,) + x[1:]
else:
x = x1 + rand_feature
if not isinstance(x, tuple):
x = (x,)
if self.pre_transform:
x = self.pre_transform(*x)
if not isinstance(x, tuple):
x = (x,)
if memory_checkpointing_enabled:
xformed = [checkpoint(t, *x) for t in self.transforms]
else:
xformed = [t(*x) for t in self.transforms]
if not isinstance(att_in, tuple):
att_in = (att_in,)
if self.feed_transforms_into_multiplexer:
att_in = att_in + (torch.stack(xformed, dim=1),)
if memory_checkpointing_enabled:
m = checkpoint(self.multiplexer, *att_in)
else:
m = self.multiplexer(*att_in)
# It is assumed that [xformed] and [m] are collapsed into tensors at this point.
outputs, attention = self.switch(xformed, m, True)
outputs = identity + outputs * self.switch_scale * fixed_scale
outputs = outputs + self.post_switch_conv(outputs) * self.psc_scale * fixed_scale
if output_attention_weights:
return outputs, attention
else:
return outputs
def set_temperature(self, temp):
self.switch.set_attention_temperature(temp)
class ConfigurableSwitchedResidualGenerator2(nn.Module):
def __init__(self, switch_depth, switch_filters, switch_reductions, switch_processing_layers, trans_counts, trans_kernel_sizes,
trans_layers, transformation_filters, attention_norm, initial_temp=20, final_temperature_step=50000, heightened_temp_min=1,
heightened_final_step=50000, upsample_factor=1,
add_scalable_noise_to_transforms=False):
super(ConfigurableSwitchedResidualGenerator2, self).__init__()
switches = []
self.initial_conv = ConvBnLelu(3, transformation_filters, norm=False, activation=False, bias=True)
self.upconv1 = ConvBnLelu(transformation_filters, transformation_filters, norm=False, bias=True)
self.upconv2 = ConvBnLelu(transformation_filters, transformation_filters, norm=False, bias=True)
self.hr_conv = ConvBnLelu(transformation_filters, transformation_filters, norm=False, bias=True)
self.final_conv = ConvBnLelu(transformation_filters, 3, norm=False, activation=False, bias=True)
for _ in range(switch_depth):
multiplx_fn = functools.partial(ConvBasisMultiplexer, transformation_filters, switch_filters, switch_reductions, switch_processing_layers, trans_counts)
pretransform_fn = functools.partial(ConvBnLelu, transformation_filters, transformation_filters, norm=False, bias=False, weight_init_factor=.1)
transform_fn = functools.partial(MultiConvBlock, transformation_filters, int(transformation_filters * 1.5), transformation_filters, kernel_size=trans_kernel_sizes, depth=trans_layers, weight_init_factor=.1)
switches.append(ConfigurableSwitchComputer(transformation_filters, multiplx_fn,
pre_transform_block=pretransform_fn, transform_block=transform_fn,
attention_norm=attention_norm,
transform_count=trans_counts, init_temp=initial_temp,
add_scalable_noise_to_transforms=add_scalable_noise_to_transforms))
self.switches = nn.ModuleList(switches)
self.transformation_counts = trans_counts
self.init_temperature = initial_temp
self.final_temperature_step = final_temperature_step
self.heightened_temp_min = heightened_temp_min
self.heightened_final_step = heightened_final_step
self.attentions = None
self.upsample_factor = upsample_factor
assert self.upsample_factor == 2 or self.upsample_factor == 4
def forward(self, x):
# This is a common bug when evaluating SRG2 generators. It needs to be configured properly in eval mode. Just fail.
if not self.train:
assert self.switches[0].switch.temperature == 1
x = self.initial_conv(x)
self.attentions = []
for i, sw in enumerate(self.switches):
x, att = sw.forward(x, True)
self.attentions.append(att)
x = self.upconv1(F.interpolate(x, scale_factor=2, mode="nearest"))
if self.upsample_factor > 2:
x = F.interpolate(x, scale_factor=2, mode="nearest")
x = self.upconv2(x)
x = self.final_conv(self.hr_conv(x))
return x, x
def set_temperature(self, temp):
[sw.set_temperature(temp) for sw in self.switches]
def update_for_step(self, step, experiments_path='.'):
if self.attentions:
temp = max(1,
1 + self.init_temperature * (self.final_temperature_step - step) / self.final_temperature_step)
if temp == 1 and self.heightened_final_step and step > self.final_temperature_step and \
self.heightened_final_step != 1:
# Once the temperature passes (1) it enters an inverted curve to match the linear curve from above.
# without this, the attention specificity "spikes" incredibly fast in the last few iterations.
h_steps_total = self.heightened_final_step - self.final_temperature_step
h_steps_current = min(step - self.final_temperature_step, h_steps_total)
# The "gap" will represent the steps that need to be traveled as a linear function.
h_gap = 1 / self.heightened_temp_min
temp = h_gap * h_steps_current / h_steps_total
# Invert temperature to represent reality on this side of the curve
temp = 1 / temp
self.set_temperature(temp)
if step % 50 == 0:
output_path = os.path.join(experiments_path, "attention_maps", "a%i")
prefix = "attention_map_%i_%%i.png" % (step,)
[save_attention_to_image_rgb(output_path % (i,), self.attentions[i], self.transformation_counts, prefix, step) for i in range(len(self.attentions))]
def get_debug_values(self, step):
temp = self.switches[0].switch.temperature
mean_hists = [compute_attention_specificity(att, 2) for att in self.attentions]
means = [i[0] for i in mean_hists]
hists = [i[1].clone().detach().cpu().flatten() for i in mean_hists]
val = {"switch_temperature": temp}
for i in range(len(means)):
val["switch_%i_specificity" % (i,)] = means[i]
val["switch_%i_histogram" % (i,)] = hists[i]
return val
# This class encapsulates an encoder based on an object detection network backbone whose purpose is to generated a
# structured embedding encoding what is in an image patch. This embedding can then be used to perform structured
# alterations to the underlying image.
#
# Caveat: Since this uses a pre-defined (and potentially pre-trained) SpineNet backbone, it has a minimum-supported
# image size, which is 128x128. In order to use 64x64 patches, you must set interpolate_first=True. though this will
# degrade quality.
class BackboneEncoder(nn.Module):
def __init__(self, interpolate_first=True, pretrained_backbone=None):
super(BackboneEncoder, self).__init__()
self.interpolate_first = interpolate_first
# Uses dual spinenets, one for the input patch and the other for the reference image.
self.patch_spine = SpineNet('49', in_channels=3, use_input_norm=True)
self.ref_spine = SpineNet('49', in_channels=3, use_input_norm=True)
self.merge_process1 = ConvGnSilu(512, 512, kernel_size=1, activation=True, norm=False, bias=True)
self.merge_process2 = ConvGnSilu(512, 384, kernel_size=1, activation=True, norm=True, bias=False)
self.merge_process3 = ConvGnSilu(384, 256, kernel_size=1, activation=False, norm=False, bias=True)
if pretrained_backbone is not None:
loaded_params = torch.load(pretrained_backbone)
self.ref_spine.load_state_dict(loaded_params['state_dict'], strict=True)
self.patch_spine.load_state_dict(loaded_params['state_dict'], strict=True)
# Returned embedding will have been reduced in size by a factor of 8 (4 if interpolate_first=True).
# Output channels are always 256.
# ex, 64x64 input with interpolate_first=True will result in tensor of shape [bx256x16x16]
def forward(self, x, ref, ref_center_point):
if self.interpolate_first:
x = F.interpolate(x, scale_factor=2, mode="bicubic")
# Don't interpolate ref - assume it is fed in at the proper resolution.
# ref = F.interpolate(ref, scale_factor=2, mode="bicubic")
# [ref] will have a 'mask' channel which we cannot use with pretrained spinenet.
ref = ref[:, :3, :, :]
ref_emb = checkpoint(self.ref_spine, ref)[0]
ref_code = gather_2d(ref_emb, ref_center_point // 8) # Divide by 8 to bring the center point to the correct location.
patch = checkpoint(self.patch_spine, x)[0]
ref_code_expanded = ref_code.view(-1, 256, 1, 1).repeat(1, 1, patch.shape[2], patch.shape[3])
combined = self.merge_process1(torch.cat([patch, ref_code_expanded], dim=1))
combined = self.merge_process2(combined)
combined = self.merge_process3(combined)
return combined
class BackboneEncoderNoRef(nn.Module):
def __init__(self, interpolate_first=True, pretrained_backbone=None):
super(BackboneEncoderNoRef, self).__init__()
self.interpolate_first = interpolate_first
self.patch_spine = SpineNet('49', in_channels=3, use_input_norm=True)
if pretrained_backbone is not None:
loaded_params = torch.load(pretrained_backbone)
self.patch_spine.load_state_dict(loaded_params['state_dict'], strict=True)
# Returned embedding will have been reduced in size by a factor of 8 (4 if interpolate_first=True).
# Output channels are always 256.
# ex, 64x64 input with interpolate_first=True will result in tensor of shape [bx256x16x16]
def forward(self, x):
if self.interpolate_first:
x = F.interpolate(x, scale_factor=2, mode="bicubic")
patch = checkpoint(self.patch_spine, x)[0]
return patch
# Note to future self:
# Can I do a real transformer here? Such as by having the multiplexer be able to toggle off of transformations by
# their output? The embedding will be used as the "Query" to the "QueryxKey=Value" relationship.
# Mutiplexer that combines a structured embedding with a contextual switch input to guide alterations to that input.
#
# Implemented as basically a u-net which reduces the input into the same structural space as the embedding, combines the
# two, then expands back into the original feature space.
class EmbeddingMultiplexer(nn.Module):
# Note: reductions=2 if the encoder is using interpolated input, otherwise reductions=3.
def __init__(self, nf, multiplexer_channels, reductions=2):
super(EmbeddingMultiplexer, self).__init__()
self.embedding_process = MultiConvBlock(256, 256, 256, kernel_size=3, depth=3, norm=True)
self.filter_conv = ConvGnSilu(nf, nf, activation=True, norm=False, bias=True)
self.reduction_blocks = nn.ModuleList([HalvingProcessingBlock(nf * 2 ** i) for i in range(reductions)])
reduction_filters = nf * 2 ** reductions
self.processing_blocks = nn.Sequential(
ConvGnSilu(reduction_filters + 256, reduction_filters + 256, kernel_size=1, activation=True, norm=False, bias=True),
ConvGnSilu(reduction_filters + 256, reduction_filters + 128, kernel_size=1, activation=True, norm=True, bias=False),
ConvGnSilu(reduction_filters + 128, reduction_filters, kernel_size=3, activation=True, norm=True, bias=False),
ConvGnSilu(reduction_filters, reduction_filters, kernel_size=3, activation=True, norm=True, bias=False))
self.expansion_blocks = nn.ModuleList([ExpansionBlock2(reduction_filters // (2 ** i)) for i in range(reductions)])
gap = nf - multiplexer_channels
cbl1_out = ((nf - (gap // 2)) // 4) * 4 # Must be multiples of 4 to use with group norm.
self.cbl1 = ConvGnSilu(nf, cbl1_out, norm=True, bias=False, num_groups=4)
cbl2_out = ((nf - (3 * gap // 4)) // 4) * 4
self.cbl2 = ConvGnSilu(cbl1_out, cbl2_out, norm=True, bias=False, num_groups=4)
self.cbl3 = ConvGnSilu(cbl2_out, multiplexer_channels, bias=True, norm=False)
def forward(self, x, embedding):
x = self.filter_conv(x)
embedding = self.embedding_process(embedding)
reduction_identities = []
for b in self.reduction_blocks:
reduction_identities.append(x)
x = b(x)
x = self.processing_blocks(torch.cat([x, embedding], dim=1))
for i, b in enumerate(self.expansion_blocks):
x = b(x, reduction_identities[-i - 1])
x = self.cbl1(x)
x = self.cbl2(x)
x = self.cbl3(x)
return x
class QueryKeyMultiplexer(nn.Module):
def __init__(self, nf, multiplexer_channels, reductions=2):
super(QueryKeyMultiplexer, self).__init__()
# Blocks used to create the query
self.input_process = ConvGnSilu(nf, nf, activation=True, norm=False, bias=True)
self.embedding_process = ConvGnSilu(256, 256, activation=True, norm=False, bias=True)
self.reduction_blocks = nn.ModuleList([HalvingProcessingBlock(nf * 2 ** i) for i in range(reductions)])
reduction_filters = nf * 2 ** reductions
self.processing_blocks = nn.Sequential(
ConvGnSilu(reduction_filters + 256, reduction_filters + 256, kernel_size=1, activation=True, norm=False, bias=True),
ConvGnSilu(reduction_filters + 256, reduction_filters + 128, kernel_size=1, activation=True, norm=True, bias=False),
ConvGnSilu(reduction_filters + 128, reduction_filters, kernel_size=3, activation=True, norm=True, bias=False),
ConvGnSilu(reduction_filters, reduction_filters, kernel_size=3, activation=True, norm=True, bias=False))
self.expansion_blocks = nn.ModuleList([ExpansionBlock2(reduction_filters // (2 ** i)) for i in range(reductions)])
# Blocks used to create the key
self.key_process = ConvGnSilu(nf, nf, kernel_size=1, activation=True, norm=False, bias=True)
# Postprocessing blocks.
self.query_key_combine = ConvGnSilu(nf*2, nf, kernel_size=1, activation=True, norm=False, bias=False)
self.cbl1 = ConvGnSilu(nf, nf // 2, kernel_size=1, norm=True, bias=False, num_groups=4)
self.cbl2 = ConvGnSilu(nf // 2, 1, kernel_size=1, norm=False, bias=False)
def forward(self, x, embedding, transformations):
q = self.input_process(x)
embedding = self.embedding_process(embedding)
reduction_identities = []
for b in self.reduction_blocks:
reduction_identities.append(q)
q = b(q)
q = self.processing_blocks(torch.cat([q, embedding], dim=1))
for i, b in enumerate(self.expansion_blocks):
q = b(q, reduction_identities[-i - 1])
b, t, f, h, w = transformations.shape
k = transformations.view(b * t, f, h, w)
k = self.key_process(k)
q = q.view(b, 1, f, h, w).repeat(1, t, 1, 1, 1).view(b * t, f, h, w)
v = self.query_key_combine(torch.cat([q, k], dim=1))
v = self.cbl1(v)
v = self.cbl2(v)
return v.view(b, t, h, w)
if __name__ == '__main__':
bb = BackboneEncoder(64)
emb = QueryKeyMultiplexer(64, 10)
x = torch.randn(4,3,64,64)
r = torch.randn(4,3,128,128)
xu = torch.randn(4,64,64,64)
cp = torch.zeros((4,2), dtype=torch.long)
trans = [torch.randn(4,64,64,64) for t in range(10)]
b = bb(x, r, cp)
emb(xu, b, trans)