280 lines
10 KiB
Python
280 lines
10 KiB
Python
import os
|
|
from time import time
|
|
|
|
import torch
|
|
import torchvision
|
|
import torch.distributed as distributed
|
|
from torch import nn
|
|
from tqdm import tqdm
|
|
|
|
from models.switched_conv.switched_conv_hard_routing import SwitchedConvHardRouting, \
|
|
convert_conv_net_state_dict_to_switched_conv
|
|
from models.vqvae.vqvae import ResBlock, Quantize
|
|
from trainer.networks import register_model
|
|
from utils.util import checkpoint, opt_get
|
|
|
|
|
|
# Upsamples and blurs (similar to StyleGAN). Replaces ConvTranspose2D from the original paper.
|
|
class UpsampleConv(nn.Module):
|
|
def __init__(self, in_filters, out_filters, kernel_size, padding, cfg):
|
|
super().__init__()
|
|
self.conv = SwitchedConvHardRouting(in_filters, out_filters, kernel_size, breadth=cfg['breadth'], include_coupler=True,
|
|
coupler_mode=cfg['mode'], coupler_dim_in=in_filters, dropout_rate=cfg['dropout'], hard_en=cfg['hard_enabled'])
|
|
|
|
def forward(self, x):
|
|
up = torch.nn.functional.interpolate(x, scale_factor=2)
|
|
return self.conv(up)
|
|
|
|
|
|
class Encoder(nn.Module):
|
|
def __init__(self, in_channel, channel, n_res_block, n_res_channel, stride, cfg):
|
|
super().__init__()
|
|
|
|
if stride == 4:
|
|
blocks = [
|
|
nn.Conv2d(in_channel, channel // 2, 5, stride=2, padding=2),
|
|
nn.LeakyReLU(inplace=True),
|
|
SwitchedConvHardRouting(channel // 2, channel, 5, breadth=cfg['breadth'], stride=2, include_coupler=True,
|
|
coupler_mode=cfg['mode'], coupler_dim_in=channel // 2, dropout_rate=cfg['dropout'], hard_en=cfg['hard_enabled']),
|
|
nn.LeakyReLU(inplace=True),
|
|
SwitchedConvHardRouting(channel, channel, 3, breadth=cfg['breadth'], include_coupler=True, coupler_mode=cfg['mode'],
|
|
coupler_dim_in=channel, dropout_rate=cfg['dropout'], hard_en=cfg['hard_enabled']),
|
|
]
|
|
|
|
elif stride == 2:
|
|
blocks = [
|
|
nn.Conv2d(in_channel, channel // 2, 5, stride=2, padding=2),
|
|
nn.LeakyReLU(inplace=True),
|
|
SwitchedConvHardRouting(channel // 2, channel, 3, breadth=cfg['breadth'], include_coupler=True, coupler_mode=cfg['mode'],
|
|
coupler_dim_in=channel // 2, dropout_rate=cfg['dropout'], hard_en=cfg['hard_enabled']),
|
|
]
|
|
|
|
for i in range(n_res_block):
|
|
blocks.append(ResBlock(channel, n_res_channel))
|
|
|
|
blocks.append(nn.LeakyReLU(inplace=True))
|
|
|
|
self.blocks = nn.Sequential(*blocks)
|
|
|
|
def forward(self, input):
|
|
return self.blocks(input)
|
|
|
|
|
|
class Decoder(nn.Module):
|
|
def __init__(
|
|
self, in_channel, out_channel, channel, n_res_block, n_res_channel, stride, cfg
|
|
):
|
|
super().__init__()
|
|
|
|
blocks = [SwitchedConvHardRouting(in_channel, channel, 3, breadth=cfg['breadth'], include_coupler=True, coupler_mode=cfg['mode'],
|
|
coupler_dim_in=in_channel, dropout_rate=cfg['dropout'], hard_en=cfg['hard_enabled'])]
|
|
|
|
for i in range(n_res_block):
|
|
blocks.append(ResBlock(channel, n_res_channel))
|
|
|
|
blocks.append(nn.LeakyReLU(inplace=True))
|
|
|
|
if stride == 4:
|
|
blocks.extend(
|
|
[
|
|
UpsampleConv(channel, channel // 2, 5, padding=2, cfg=cfg),
|
|
nn.LeakyReLU(inplace=True),
|
|
UpsampleConv(
|
|
channel // 2, out_channel, 5, padding=2, cfg=cfg
|
|
),
|
|
]
|
|
)
|
|
|
|
elif stride == 2:
|
|
blocks.append(
|
|
UpsampleConv(channel, out_channel, 5, padding=2, cfg=cfg)
|
|
)
|
|
|
|
self.blocks = nn.Sequential(*blocks)
|
|
|
|
def forward(self, input):
|
|
return self.blocks(input)
|
|
|
|
|
|
class VQVAE3HardSwitch(nn.Module):
|
|
def __init__(
|
|
self,
|
|
in_channel=3,
|
|
channel=128,
|
|
n_res_block=2,
|
|
n_res_channel=32,
|
|
codebook_dim=64,
|
|
codebook_size=512,
|
|
decay=0.99,
|
|
cfg={'mode':'standard', 'breadth':16, 'hard_enabled': True, 'dropout': 0.4}
|
|
):
|
|
super().__init__()
|
|
|
|
self.cfg = cfg
|
|
self.initial_conv = nn.Sequential(*[nn.Conv2d(in_channel, 32, 3, padding=1),
|
|
nn.LeakyReLU(inplace=True)])
|
|
self.enc_b = Encoder(32, channel, n_res_block, n_res_channel, stride=4, cfg=cfg)
|
|
self.enc_t = Encoder(channel, channel, n_res_block, n_res_channel, stride=2, cfg=cfg)
|
|
self.quantize_conv_t = nn.Conv2d(channel, codebook_dim, 1)
|
|
self.quantize_t = Quantize(codebook_dim, codebook_size)
|
|
self.dec_t = Decoder(
|
|
codebook_dim, codebook_dim, channel, n_res_block, n_res_channel, stride=2, cfg=cfg
|
|
)
|
|
self.quantize_conv_b = nn.Conv2d(codebook_dim + channel, codebook_dim, 1)
|
|
self.quantize_b = Quantize(codebook_dim, codebook_size)
|
|
self.upsample_t = UpsampleConv(
|
|
codebook_dim, codebook_dim, 5, padding=2, cfg=cfg
|
|
)
|
|
self.dec = Decoder(
|
|
codebook_dim + codebook_dim,
|
|
32,
|
|
channel,
|
|
n_res_block,
|
|
n_res_channel,
|
|
stride=4,
|
|
cfg=cfg
|
|
)
|
|
self.final_conv = nn.Conv2d(32, in_channel, 3, padding=1)
|
|
|
|
def forward(self, input):
|
|
quant_t, quant_b, diff, _, _ = self.encode(input)
|
|
dec = self.decode(quant_t, quant_b)
|
|
|
|
return dec, diff
|
|
|
|
def save_attention_to_image_rgb(self, output_file, attention_out, attention_size, cmap_discrete_name='viridis'):
|
|
from matplotlib import cm
|
|
magnitude, indices = torch.topk(attention_out, 3, dim=1)
|
|
indices = indices.cpu()
|
|
colormap = cm.get_cmap(cmap_discrete_name, attention_size)
|
|
img = torch.tensor(colormap(indices[:, 0, :, :].detach().numpy())) # TODO: use other k's
|
|
img = img.permute((0, 3, 1, 2))
|
|
torchvision.utils.save_image(img, output_file)
|
|
|
|
def visual_dbg(self, step, path):
|
|
convs = [self.dec.blocks[-1].conv, self.dec_t.blocks[-1].conv, self.enc_b.blocks[-4], self.enc_t.blocks[-4]]
|
|
for i, c in enumerate(convs):
|
|
self.save_attention_to_image_rgb(os.path.join(path, "%i_selector_%i.png" % (step, i+1)), c.last_select, 16)
|
|
|
|
def get_debug_values(self, step, __):
|
|
switched_convs = [('enc_b_blk2', self.enc_b.blocks[2]),
|
|
('enc_b_blk4', self.enc_b.blocks[4]),
|
|
('enc_t_blk2', self.enc_t.blocks[2]),
|
|
('dec_t_blk0', self.dec_t.blocks[0]),
|
|
('dec_t_blk-1', self.dec_t.blocks[-1].conv),
|
|
('dec_blk0', self.dec.blocks[0]),
|
|
('dec_blk-1', self.dec.blocks[-1].conv),
|
|
('dec_blk-3', self.dec.blocks[-3].conv)]
|
|
logs = {}
|
|
for name, swc in switched_convs:
|
|
logs[f'{name}_histogram_switch_usage'] = swc.latest_masks
|
|
return logs
|
|
|
|
def encode(self, input):
|
|
fea = self.initial_conv(input)
|
|
enc_b = checkpoint(self.enc_b, fea)
|
|
enc_t = checkpoint(self.enc_t, enc_b)
|
|
|
|
quant_t = self.quantize_conv_t(enc_t).permute(0, 2, 3, 1)
|
|
quant_t, diff_t, id_t = self.quantize_t(quant_t)
|
|
quant_t = quant_t.permute(0, 3, 1, 2)
|
|
diff_t = diff_t.unsqueeze(0)
|
|
|
|
dec_t = checkpoint(self.dec_t, quant_t)
|
|
enc_b = torch.cat([dec_t, enc_b], 1)
|
|
|
|
quant_b = checkpoint(self.quantize_conv_b, enc_b).permute(0, 2, 3, 1)
|
|
quant_b, diff_b, id_b = self.quantize_b(quant_b)
|
|
quant_b = quant_b.permute(0, 3, 1, 2)
|
|
diff_b = diff_b.unsqueeze(0)
|
|
|
|
return quant_t, quant_b, diff_t + diff_b, id_t, id_b
|
|
|
|
def decode(self, quant_t, quant_b):
|
|
upsample_t = self.upsample_t(quant_t)
|
|
quant = torch.cat([upsample_t, quant_b], 1)
|
|
dec = checkpoint(self.dec, quant)
|
|
dec = checkpoint(self.final_conv, dec)
|
|
|
|
return dec
|
|
|
|
def decode_code(self, code_t, code_b):
|
|
quant_t = self.quantize_t.embed_code(code_t)
|
|
quant_t = quant_t.permute(0, 3, 1, 2)
|
|
quant_b = self.quantize_b.embed_code(code_b)
|
|
quant_b = quant_b.permute(0, 3, 1, 2)
|
|
|
|
dec = self.decode(quant_t, quant_b)
|
|
|
|
return dec
|
|
|
|
|
|
|
|
def convert_weights(weights_file):
|
|
sd = torch.load(weights_file)
|
|
from models.vqvae.vqvae_3 import VQVAE3
|
|
std_model = VQVAE3()
|
|
std_model.load_state_dict(sd)
|
|
nsd = convert_conv_net_state_dict_to_switched_conv(std_model, 16, ['quantize_conv_t', 'quantize_conv_b',
|
|
'enc_b.blocks.0', 'enc_t.blocks.0',
|
|
'conv.1', 'conv.3', 'initial_conv', 'final_conv'])
|
|
torch.save(nsd, "converted.pth")
|
|
|
|
|
|
@register_model
|
|
def register_vqvae3_hard_switch(opt_net, opt):
|
|
kw = opt_get(opt_net, ['kwargs'], {})
|
|
vq = VQVAE3HardSwitch(**kw)
|
|
if distributed.is_initialized() and distributed.get_world_size() > 1:
|
|
vq = torch.nn.SyncBatchNorm.convert_sync_batchnorm(vq)
|
|
return vq
|
|
|
|
|
|
def performance_test():
|
|
# For breadth=32:
|
|
# Custom_cuda_naive: 15.4
|
|
# Torch_native: 29.2s
|
|
#
|
|
# For breadth=8
|
|
# Custom_cuda_naive: 9.8
|
|
# Torch_native: 10s
|
|
cfg = {
|
|
'mode': 'lambda',
|
|
'breadth': 16,
|
|
'hard_enabled': True,
|
|
'dropout': 0,
|
|
}
|
|
net = VQVAE3HardSwitch(cfg=cfg).to('cuda').double()
|
|
cfg['hard_enabled'] = False
|
|
netO = VQVAE3HardSwitch(cfg=cfg).double()
|
|
netO.load_state_dict(net.state_dict())
|
|
netO = netO.cpu()
|
|
|
|
loss = nn.L1Loss()
|
|
opt = torch.optim.Adam(net.parameters(), lr=1e-4)
|
|
started = time()
|
|
for j in tqdm(range(10)):
|
|
inp = torch.rand((4, 3, 64, 64), device='cuda', dtype=torch.double)
|
|
res = net(inp)[0]
|
|
l = loss(res, inp)
|
|
l.backward()
|
|
|
|
res2 = netO(inp.cpu())[0]
|
|
l = loss(res2, inp.cpu())
|
|
l.backward()
|
|
|
|
for p, op in zip(net.parameters(), netO.parameters()):
|
|
diff = p.grad.cpu() - op.grad
|
|
print(diff.max())
|
|
|
|
opt.step()
|
|
net.zero_grad()
|
|
print("Elapsed: ", (time()-started))
|
|
|
|
|
|
if __name__ == '__main__':
|
|
#v = VQVAE3HardSwitch()
|
|
#print(v(torch.randn(1,3,128,128))[0].shape)
|
|
#convert_weights("../../../experiments/vqvae_base.pth")
|
|
performance_test()
|