diff --git a/codes/models/switched_conv/switched_conv_hard_routing.py b/codes/models/switched_conv/switched_conv_hard_routing.py index 5e590ac7..d87d4909 100644 --- a/codes/models/switched_conv/switched_conv_hard_routing.py +++ b/codes/models/switched_conv/switched_conv_hard_routing.py @@ -146,14 +146,16 @@ class SwitchNorm(nn.Module): class HardRoutingGate(nn.Module): - def __init__(self, breadth): + def __init__(self, breadth, hard_en=True): super().__init__() self.norm = SwitchNorm(breadth, accumulator_size=256) + self.hard_en = hard_en def forward(self, x): soft = self.norm(nn.functional.softmax(x, dim=1)) - hard = RouteTop1.apply(soft) # This variant can route gradients downstream. - return hard + if self.hard_en: + return RouteTop1.apply(soft) + return soft class SwitchedConvHardRouting(nn.Module): @@ -167,7 +169,8 @@ class SwitchedConvHardRouting(nn.Module): dropout_rate=0.0, include_coupler: bool = False, # A 'coupler' is a latent converter which can make any bxcxhxw tensor a compatible switchedconv selector by performing a linear 1x1 conv, softmax and interpolate. coupler_mode: str = 'standard', - coupler_dim_in: int = 0): + coupler_dim_in: int = 0, + hard_en=True): # A test switch that, when used in 'emulation mode' (where all convs are calculated using torch functions) computes soft-attention instead of hard-attention. super().__init__() self.in_channels = in_c self.out_channels = out_c @@ -190,7 +193,7 @@ class SwitchedConvHardRouting(nn.Module): Conv2d(breadth, breadth, 1, stride=self.stride)) else: self.coupler = None - self.gate = HardRoutingGate(breadth) + self.gate = HardRoutingGate(breadth, hard_en=hard_en) self.weight = nn.Parameter(torch.empty(out_c, in_c, breadth, kernel_sz, kernel_sz)) if bias: diff --git a/codes/models/vqvae/vqvae_3_hardswitch.py b/codes/models/vqvae/vqvae_3_hardswitch.py index 958eda98..212e912c 100644 --- a/codes/models/vqvae/vqvae_3_hardswitch.py +++ b/codes/models/vqvae/vqvae_3_hardswitch.py @@ -15,10 +15,10 @@ 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): + def __init__(self, in_filters, out_filters, kernel_size, padding, cfg): super().__init__() - self.conv = SwitchedConvHardRouting(in_filters, out_filters, kernel_size, breadth=16, include_coupler=True, - coupler_mode='standard', coupler_dim_in=in_filters, dropout_rate=0.4) + 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) @@ -26,26 +26,26 @@ class UpsampleConv(nn.Module): class Encoder(nn.Module): - def __init__(self, in_channel, channel, n_res_block, n_res_channel, stride): + 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=16, stride=2, include_coupler=True, - coupler_mode='standard', coupler_dim_in=channel // 2, dropout_rate=0.4), + 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=16, include_coupler=True, coupler_mode='standard', - coupler_dim_in=channel, dropout_rate=0.4), + 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=16, include_coupler=True, coupler_mode='standard', - coupler_dim_in=channel // 2, dropout_rate=0.4), + 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): @@ -61,12 +61,12 @@ class Encoder(nn.Module): class Decoder(nn.Module): def __init__( - self, in_channel, out_channel, channel, n_res_block, n_res_channel, stride + self, in_channel, out_channel, channel, n_res_block, n_res_channel, stride, cfg ): super().__init__() - blocks = [SwitchedConvHardRouting(in_channel, channel, 3, breadth=16, include_coupler=True, coupler_mode='standard', - coupler_dim_in=in_channel, dropout_rate=0.4)] + 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)) @@ -76,17 +76,17 @@ class Decoder(nn.Module): if stride == 4: blocks.extend( [ - UpsampleConv(channel, channel // 2, 5, padding=2), + UpsampleConv(channel, channel // 2, 5, padding=2, cfg=cfg), nn.LeakyReLU(inplace=True), UpsampleConv( - channel // 2, out_channel, 5, padding=2 + channel // 2, out_channel, 5, padding=2, cfg=cfg ), ] ) elif stride == 2: blocks.append( - UpsampleConv(channel, out_channel, 5, padding=2) + UpsampleConv(channel, out_channel, 5, padding=2, cfg=cfg) ) self.blocks = nn.Sequential(*blocks) @@ -105,22 +105,24 @@ class VQVAE3HardSwitch(nn.Module): 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) - self.enc_t = Encoder(channel, channel, n_res_block, n_res_channel, stride=2) + 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 + 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 + codebook_dim, codebook_dim, 5, padding=2, cfg=cfg ) self.dec = Decoder( codebook_dim + codebook_dim, @@ -129,6 +131,7 @@ class VQVAE3HardSwitch(nn.Module): n_res_block, n_res_channel, stride=4, + cfg=cfg ) self.final_conv = nn.Conv2d(32, in_channel, 3, padding=1) @@ -211,7 +214,7 @@ def convert_weights(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', + nsd = convert_conv_net_state_dict_to_switched_conv(std_model, 8, ['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") @@ -224,7 +227,13 @@ def register_vqvae3_hard_switch(opt_net, opt): def performance_test(): - net = VQVAE3HardSwitch().to('cuda') + cfg = { + 'mode': 'lambda', + 'breadth': 8, + 'hard_enabled': False, + 'dropout': 0 + } + net = VQVAE3HardSwitch(cfg=cfg).to('cuda') loss = nn.L1Loss() opt = torch.optim.Adam(net.parameters(), lr=1e-4) started = time() @@ -241,5 +250,5 @@ def performance_test(): if __name__ == '__main__': #v = VQVAE3HardSwitch() #print(v(torch.randn(1,3,128,128))[0].shape) - #convert_weights("../../../experiments/test_vqvae3.pth") - performance_test() + convert_weights("../../../experiments/test_vqvae3.pth") + #performance_test() diff --git a/codes/train.py b/codes/train.py index 0a0bf319..515ea149 100644 --- a/codes/train.py +++ b/codes/train.py @@ -295,7 +295,7 @@ class Trainer: if __name__ == '__main__': parser = argparse.ArgumentParser() - parser.add_argument('-opt', type=str, help='Path to option YAML file.', default='../options/train_imgset_tiled_nvqvae_stage1_lambda.yml') + parser.add_argument('-opt', type=str, help='Path to option YAML file.', default='../options/train_imgset_vqvae3_stage1.yml') parser.add_argument('--launcher', choices=['none', 'pytorch'], default='none', help='job launcher') parser.add_argument('--local_rank', type=int, default=0) args = parser.parse_args()