From 64bfcc83637d236477303bfad0cd598f96d5e023 Mon Sep 17 00:00:00 2001 From: logtd Date: Sat, 24 Aug 2024 17:24:55 -0600 Subject: [PATCH] init --- .gitignore | 3 + __init__.py | 16 + modules/configs.py | 56 +++ modules/seecoder.py | 520 +++++++++++++++++++++++ modules/swin.py | 657 +++++++++++++++++++++++++++++ nodes/load_seecoder_node.py | 36 ++ nodes/load_seecoder_uncond_node.py | 25 ++ nodes/seecoder_encode_node.py | 29 ++ 8 files changed, 1342 insertions(+) create mode 100644 .gitignore create mode 100644 __init__.py create mode 100644 modules/configs.py create mode 100644 modules/seecoder.py create mode 100644 modules/swin.py create mode 100644 nodes/load_seecoder_node.py create mode 100644 nodes/load_seecoder_uncond_node.py create mode 100644 nodes/seecoder_encode_node.py diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..65a35fb --- /dev/null +++ b/.gitignore @@ -0,0 +1,3 @@ +**/__pycache__ +**/*.pyc +.DS_Store \ No newline at end of file diff --git a/__init__.py b/__init__.py new file mode 100644 index 0000000..b5ea0e2 --- /dev/null +++ b/__init__.py @@ -0,0 +1,16 @@ +from .nodes.load_seecoder_node import LoadSeeCoderNode +from .nodes.seecoder_encode_node import SeecoderEncodeNode +from .nodes.load_seecoder_uncond_node import LoadSeeCoderUncondNode + + +NODE_CLASS_MAPPINGS = { + 'LoadSeeCoder': LoadSeeCoderNode, + 'SeecoderEncode': SeecoderEncodeNode, + 'LoadSeeCoderUncond': LoadSeeCoderUncondNode, +} + +NODE_DISPLAY_NAME_MAPPINGS = { + 'LoadSeeCoder': 'Seecoder Load Model', + 'SeecoderEncode': 'Seecoder Encode Latent', + 'LoadSeeCoderUncond': 'Seecoder Load Uncond', +} diff --git a/modules/configs.py b/modules/configs.py new file mode 100644 index 0000000..14814c5 --- /dev/null +++ b/modules/configs.py @@ -0,0 +1,56 @@ + + + +seecoder_decoder_config = { + 'inchannels': { + 'res3': 384, + 'res4': 768, + 'res5': 1536 + }, + 'trans_input_tags': ['res3', 'res4', 'res5'], + 'trans_dim': 768, + 'trans_dropout': 0.1, + 'trans_nheads': 8, + 'trans_feedforward_dim': 1024, + 'trans_num_layers': 6 +} + +seecoder_query_config = { + 'in_channels' : 768, + 'hidden_dim': 768, + 'num_queries': [4, 144], + 'nheads': 8, + 'num_layers': 9, + 'feedforward_dim': 2048, + 'pre_norm': False, + 'num_feature_levels': 3, + 'enforce_input_project': False, + 'with_fea2d_pos': False +} + +# Not used +seecoder_query_pa_config = { + **seecoder_query_config, + 'with_fea2d_pos': True +} + +swin_config = { + 'embed_dim': 128, + 'depths': [ 2, 2, 18, 2 ], + 'num_heads': [ 4, 8, 16, 32 ], + 'window_size': 7, + 'ape': False, + 'drop_path_rate': 0.3, + 'patch_norm': True, +} + +swin_large_config = { + **swin_config, + 'embed_dim': 192, + 'depths': [ 2, 2, 18, 2 ], + 'num_heads': [ 6, 12, 24, 48 ], + 'window_size': 12, + 'ape': False, + 'drop_path_rate': 0.3, + 'patch_norm': True, +} \ No newline at end of file diff --git a/modules/seecoder.py b/modules/seecoder.py new file mode 100644 index 0000000..dc9ca44 --- /dev/null +++ b/modules/seecoder.py @@ -0,0 +1,520 @@ +import torch +import torch.nn as nn +import torch.nn.functional as F +import copy + + +from .configs import swin_large_config, seecoder_decoder_config, seecoder_query_config, seecoder_query_pa_config +from .swin import SwinTransformer + +########### +# helpers # +########### + +def with_pos_embed(x, pos): + return x if pos is None else x + pos + +def _get_clones(module, N): + return nn.ModuleList([copy.deepcopy(module) for i in range(N)]) + +def _get_activation_fn(activation): + """Return an activation function given a string""" + if activation == "relu": + return F.relu + if activation == "gelu": + return F.gelu + if activation == "glu": + return F.glu + raise RuntimeError(f"activation should be relu/gelu, not {activation}.") + +def c2_xavier_fill(module): + # Caffe2 implementation of XavierFill in fact + nn.init.kaiming_uniform_(module.weight, a=1) + if module.bias is not None: + nn.init.constant_(module.bias, 0) + +def with_pos_embed(x, pos): + return x if pos is None else x + pos + +########### +# Modules # +########### + +class Conv2d_Convenience(nn.Conv2d): + def __init__(self, *args, **kwargs): + norm = kwargs.pop("norm", None) + activation = kwargs.pop("activation", None) + super().__init__(*args, **kwargs) + self.norm = norm + self.activation = activation + + def forward(self, x): + x = F.conv2d( + x, self.weight, self.bias, self.stride, self.padding, self.dilation, self.groups) + if self.norm is not None: + x = self.norm(x) + if self.activation is not None: + x = self.activation(x) + return x + + +class DecoderLayer(nn.Module): + def __init__(self, + dim=256, + feedforward_dim=1024, + dropout=0.1, + activation="relu", + n_heads=8,): + + super().__init__() + + self.self_attn = nn.MultiheadAttention(dim, n_heads, dropout=dropout) + self.dropout1 = nn.Dropout(dropout) + self.norm1 = nn.LayerNorm(dim) + + self.linear1 = nn.Linear(dim, feedforward_dim) + self.activation = _get_activation_fn(activation) + self.dropout2 = nn.Dropout(dropout) + self.linear2 = nn.Linear(feedforward_dim, dim) + self.dropout3 = nn.Dropout(dropout) + self.norm2 = nn.LayerNorm(dim) + + def forward(self, x): + h = x + h1 = self.self_attn(x, x, x, attn_mask=None)[0] + h = h + self.dropout1(h1) + h = self.norm1(h) + + h2 = self.linear2(self.dropout2(self.activation(self.linear1(h)))) + h = h + self.dropout3(h2) + h = self.norm2(h) + return h + +class DecoderLayerStacked(nn.Module): + def __init__(self, layer, num_layers, norm=None): + super().__init__() + self.layers = _get_clones(layer, num_layers) + self.num_layers = num_layers + self.norm = norm + + def forward(self, x): + h = x + for _, layer in enumerate(self.layers): + h = layer(h) + if self.norm is not None: + h = self.norm(h) + return h + +class SelfAttentionLayer(nn.Module): + def __init__(self, channels, nhead, dropout=0.0, + activation="relu", normalize_before=False): + super().__init__() + self.self_attn = nn.MultiheadAttention(channels, nhead, dropout=dropout) + + self.norm = nn.LayerNorm(channels) + self.dropout = nn.Dropout(dropout) + + self.activation = _get_activation_fn(activation) + self.normalize_before = normalize_before + + def forward(self, + qkv, + qk_pos = None, + mask = None,): + h = qkv + qk = with_pos_embed(qkv, qk_pos).transpose(0, 1) + v = qkv.transpose(0, 1) + h1 = self.self_attn(qk, qk, v, attn_mask=mask)[0] + h1 = h1.transpose(0, 1) + h = h + self.dropout(h1) + h = self.norm(h) + return h + + +class CrossAttentionLayer(nn.Module): + def __init__(self, channels, nhead, dropout=0.0, + activation="relu", normalize_before=False): + super().__init__() + self.multihead_attn = nn.MultiheadAttention(channels, nhead, dropout=dropout) + + self.norm = nn.LayerNorm(channels) + self.dropout = nn.Dropout(dropout) + + self.activation = _get_activation_fn(activation) + self.normalize_before = normalize_before + + def forward(self, + q, + kv, + q_pos = None, + k_pos = None, + mask = None,): + h = q + q = with_pos_embed(q, q_pos).transpose(0, 1) + k = with_pos_embed(kv, k_pos).transpose(0, 1) + v = kv.transpose(0, 1) + h1 = self.multihead_attn(q, k, v, attn_mask=mask)[0] + h1 = h1.transpose(0, 1) + h = h + self.dropout(h1) + h = self.norm(h) + return h + + +class FeedForwardLayer(nn.Module): + def __init__(self, channels, hidden_channels=2048, dropout=0.0, + activation="relu", normalize_before=False): + super().__init__() + self.linear1 = nn.Linear(channels, hidden_channels) + self.dropout = nn.Dropout(dropout) + self.linear2 = nn.Linear(hidden_channels, channels) + self.norm = nn.LayerNorm(channels) + self.activation = _get_activation_fn(activation) + self.normalize_before = normalize_before + + def forward_post(self, x): + h = x + h1 = self.linear2(self.dropout(self.activation(self.linear1(h)))) + h = h + self.dropout(h1) + h = self.norm(h) + return h + + def forward_pre(self, x): + xn = self.norm(x) + h = x + h1 = self.linear2(self.dropout(self.activation(self.linear1(xn)))) + h = h + self.dropout(h1) + return h + + def forward(self, *args, **kwargs): + if self.normalize_before: + return self.forward_pre(*args, **kwargs) + return self.forward_post(*args, **kwargs) + + +class MLP(nn.Module): + def __init__(self, in_channels, channels, out_channels, num_layers): + super().__init__() + self.num_layers = num_layers + h = [channels] * (num_layers - 1) + self.layers = nn.ModuleList( + nn.Linear(n, k) + for n, k in zip([in_channels]+h, h+[out_channels])) + + def forward(self, x): + for i, layer in enumerate(self.layers): + x = F.relu(layer(x)) if i < self.num_layers - 1 else layer(x) + return x + + +class PPE_MLP(nn.Module): + def __init__(self, freq_num=20, freq_max=None, out_channel=768, mlp_layer=3): + import math + super().__init__() + self.freq_num = freq_num + self.freq_max = freq_max + self.out_channel = out_channel + self.mlp_layer = mlp_layer + self.twopi = 2 * math.pi + + mlp = [] + in_channel = freq_num*4 + for idx in range(mlp_layer): + linear = nn.Linear(in_channel, out_channel, bias=True) + nn.init.xavier_normal_(linear.weight) + nn.init.constant_(linear.bias, 0) + mlp.append(linear) + if idx != mlp_layer-1: + mlp.append(nn.SiLU()) + in_channel = out_channel + self.mlp = nn.Sequential(*mlp) + nn.init.constant_(self.mlp[-1].weight, 0) + + def forward(self, x, mask=None): + assert mask is None, "Mask not implemented" + h, w = x.shape[-2:] + minlen = min(h, w) + + h_embed, w_embed = torch.meshgrid(torch.arange(h), torch.arange(w), indexing='ij') + if self.training: + import numpy.random as npr + pertube_h, pertube_w = npr.uniform(-0.5, 0.5), npr.uniform(-0.5, 0.5) + else: + pertube_h, pertube_w = 0, 0 + + h_embed = (h_embed+0.5 - h/2 + pertube_h) / (minlen) * self.twopi + w_embed = (w_embed+0.5 - w/2 + pertube_w) / (minlen) * self.twopi + h_embed, w_embed = h_embed.to(x.device).to(x.dtype), w_embed.to(x.device).to(x.dtype) + + dim_t = torch.linspace(0, 1, self.freq_num, dtype=torch.float32, device=x.device) + freq_max = self.freq_max if self.freq_max is not None else minlen/2 + dim_t = freq_max ** dim_t.to(x.dtype) + + pos_h = h_embed[:, :, None] * dim_t + pos_w = w_embed[:, :, None] * dim_t + pos = torch.cat((pos_h.sin(), pos_h.cos(), pos_w.sin(), pos_w.cos()), dim=-1) + pos = self.mlp(pos) + pos = pos.permute(2, 0, 1)[None] + return pos + + def __repr__(self, _repr_indent=4): + head = "Positional encoding " + self.__class__.__name__ + body = [ + "num_pos_feats: {}".format(self.num_pos_feats), + "temperature: {}".format(self.temperature), + "normalize: {}".format(self.normalize), + "scale: {}".format(self.scale), + ] + # _repr_indent = 4 + lines = [head] + [" " * _repr_indent + line for line in body] + return "\n".join(lines) + +########### +# Decoder # +########### + +class Decoder(nn.Module): + def __init__( + self, + inchannels, + trans_input_tags, + trans_num_layers, + trans_dim, + trans_nheads, + trans_dropout, + trans_feedforward_dim,): + + super().__init__() + trans_inchannels = { + k: v for k, v in inchannels.items() if k in trans_input_tags} + fpn_inchannels = { + k: v for k, v in inchannels.items() if k not in trans_input_tags} + + self.trans_tags = sorted(list(trans_inchannels.keys())) + self.fpn_tags = sorted(list(fpn_inchannels.keys())) + self.all_tags = sorted(list(inchannels.keys())) + + if len(self.trans_tags)==0: + assert False # Not allowed + + self.num_trans_lvls = len(self.trans_tags) + + self.inproj_layers = nn.ModuleDict() + for tagi in self.trans_tags: + layeri = nn.Sequential( + nn.Conv2d(trans_inchannels[tagi], trans_dim, kernel_size=1), + nn.GroupNorm(32, trans_dim),) + nn.init.xavier_uniform_(layeri[0].weight, gain=1) + nn.init.constant_(layeri[0].bias, 0) + self.inproj_layers[tagi] = layeri + + tlayer = DecoderLayer( + dim = trans_dim, + n_heads = trans_nheads, + dropout = trans_dropout, + feedforward_dim = trans_feedforward_dim, + activation = 'relu',) + + self.transformer = DecoderLayerStacked(tlayer, trans_num_layers) + for p in self.transformer.parameters(): + if p.dim() > 1: + nn.init.xavier_uniform_(p) + self.level_embed = nn.Parameter(torch.Tensor(len(self.trans_tags), trans_dim)) + nn.init.normal_(self.level_embed) + + self.lateral_layers = nn.ModuleDict() + self.output_layers = nn.ModuleDict() + for tagi in self.all_tags: + lateral_conv = Conv2d_Convenience( + inchannels[tagi], trans_dim, kernel_size=1, + bias=False, norm=nn.GroupNorm(32, trans_dim)) + c2_xavier_fill(lateral_conv) + self.lateral_layers[tagi] = lateral_conv + + for tagi in self.fpn_tags: + output_conv = Conv2d_Convenience( + trans_dim, trans_dim, kernel_size=3, stride=1, padding=1, + bias=False, norm=nn.GroupNorm(32, trans_dim), activation=F.relu,) + c2_xavier_fill(output_conv) + self.output_layers[tagi] = output_conv + + def forward(self, features): + x = [] + spatial_shapes = {} + for idx, tagi in enumerate(self.trans_tags[::-1]): + xi = features[tagi] + xi = self.inproj_layers[tagi](xi) + bs, _, h, w = xi.shape + spatial_shapes[tagi] = (h, w) + xi = xi.flatten(2).transpose(1, 2) + self.level_embed[idx].view(1, 1, -1) + x.append(xi) + + x_length = [xi.shape[1] for xi in x] + x_concat = torch.cat(x, 1) + y_concat = self.transformer(x_concat) + y = torch.split(y_concat, x_length, dim=1) + + out = {} + for idx, tagi in enumerate(self.trans_tags[::-1]): + h, w = spatial_shapes[tagi] + yi = y[idx].transpose(1, 2).view(bs, -1, h, w) + out[tagi] = yi + + for idx, tagi in enumerate(self.all_tags[::-1]): + lconv = self.lateral_layers[tagi] + if tagi in self.trans_tags: + out[tagi] = out[tagi] + lconv(features[tagi]) + tag_save = tagi + else: + oconv = self.output_layers[tagi] + h = lconv(features[tagi]) + oprev = out[tag_save] + h = h + F.interpolate(oconv(oprev), size=h.shape[-2:], mode="bilinear", align_corners=False) + out[tagi] = h + + return out + +##################### +# Query Transformer # +##################### + +class QueryTransformer(nn.Module): + def __init__(self, + in_channels, + hidden_dim, + num_queries = [8, 144], + nheads = 8, + num_layers = 9, + feedforward_dim = 2048, + mask_dim = 256, + pre_norm = False, + num_feature_levels = 3, + enforce_input_project = False, + with_fea2d_pos = True): + + super().__init__() + + if with_fea2d_pos: + self.pe_layer = PPE_MLP(freq_num=20, freq_max=None, out_channel=hidden_dim, mlp_layer=3) + else: + self.pe_layer = None + + if in_channels!=hidden_dim or enforce_input_project: + self.input_proj = nn.ModuleList() + for _ in range(num_feature_levels): + self.input_proj.append(nn.Conv2d(in_channels, hidden_dim, kernel_size=1)) + c2_xavier_fill(self.input_proj[-1]) + else: + self.input_proj = None + + self.num_heads = nheads + self.num_layers = num_layers + self.transformer_selfatt_layers = nn.ModuleList() + self.transformer_crossatt_layers = nn.ModuleList() + self.transformer_feedforward_layers = nn.ModuleList() + + for _ in range(self.num_layers): + self.transformer_selfatt_layers.append( + SelfAttentionLayer( + channels=hidden_dim, + nhead=nheads, + dropout=0.0, + normalize_before=pre_norm, )) + + self.transformer_crossatt_layers.append( + CrossAttentionLayer( + channels=hidden_dim, + nhead=nheads, + dropout=0.0, + normalize_before=pre_norm, )) + + self.transformer_feedforward_layers.append( + FeedForwardLayer( + channels=hidden_dim, + hidden_channels=feedforward_dim, + dropout=0.0, + normalize_before=pre_norm, )) + + self.num_queries = num_queries + num_gq, num_lq = self.num_queries + self.init_query = nn.Embedding(num_gq+num_lq, hidden_dim) + self.query_pos_embedding = nn.Embedding(num_gq+num_lq, hidden_dim) + + self.num_feature_levels = num_feature_levels + self.level_embed = nn.Embedding(num_feature_levels, hidden_dim) + + def forward(self, x): + # x is a list of multi-scale feature + assert len(x) == self.num_feature_levels + fea2d = [] + fea2d_pos = [] + size_list = [] + + for i in range(self.num_feature_levels): + size_list.append(x[i].shape[-2:]) + if self.pe_layer is not None: + pi = self.pe_layer(x[i], None).flatten(2) + pi = pi.transpose(1, 2) + else: + pi = None + xi = self.input_proj[i](x[i]) if self.input_proj is not None else x[i] + xi = xi.flatten(2) + self.level_embed.weight[i][None, :, None] + xi = xi.transpose(1, 2) + fea2d.append(xi) + fea2d_pos.append(pi) + + bs, _, _ = fea2d[0].shape + num_gq, num_lq = self.num_queries + gquery = self.init_query.weight[:num_gq].unsqueeze(0).repeat(bs, 1, 1) + lquery = self.init_query.weight[num_gq:].unsqueeze(0).repeat(bs, 1, 1) + + gquery_pos = self.query_pos_embedding.weight[:num_gq].unsqueeze(0).repeat(bs, 1, 1) + lquery_pos = self.query_pos_embedding.weight[num_gq:].unsqueeze(0).repeat(bs, 1, 1) + + for i in range(self.num_layers): + level_index = i % self.num_feature_levels + + qout = self.transformer_crossatt_layers[i]( + q = lquery, + kv = fea2d[level_index], + q_pos = lquery_pos, + k_pos = fea2d_pos[level_index], + mask = None,) + lquery = qout + + qout = self.transformer_selfatt_layers[i]( + qkv = torch.cat([gquery, lquery], dim=1), + qk_pos = torch.cat([gquery_pos, lquery_pos], dim=1),) + + qout = self.transformer_feedforward_layers[i](qout) + + gquery = qout[:, :num_gq] + lquery = qout[:, num_gq:] + + output = torch.cat([gquery, lquery], dim=1) + + return output + +################## +# Main structure # +################## + +class SemanticContextEncoder(nn.Module): + def __init__(self, has_pe=False): + super().__init__() + query_config = seecoder_query_config if not has_pe else seecoder_query_pa_config + self.imencoder = SwinTransformer(**swin_large_config) + self.imdecoder = Decoder(**seecoder_decoder_config) + self.qtransformer = QueryTransformer(**query_config) + + def forward(self, x): + fea = self.imencoder(x) + hs = {'res3' : fea['res3'], + 'res4' : fea['res4'], + 'res5' : fea['res5'], } + hs = self.imdecoder(hs) + hs = [hs['res3'], hs['res4'], hs['res5']] + q = self.qtransformer(hs) + return q + + def encode(self, x): + return self(x) \ No newline at end of file diff --git a/modules/swin.py b/modules/swin.py new file mode 100644 index 0000000..bbfb25c --- /dev/null +++ b/modules/swin.py @@ -0,0 +1,657 @@ +import torch +import torch.nn as nn +import torch.nn.functional as F +import torch.utils.checkpoint as checkpoint +import numpy as np + + +############################## +# timm.models.layers helpers # +############################## + +def drop_path(x, drop_prob: float = 0., training: bool = False, scale_by_keep: bool = True): + if drop_prob == 0. or not training: + return x + keep_prob = 1 - drop_prob + shape = (x.shape[0],) + (1,) * (x.ndim - 1) # work with diff dim tensors, not just 2D ConvNets + random_tensor = x.new_empty(shape).bernoulli_(keep_prob) + if keep_prob > 0.0 and scale_by_keep: + random_tensor.div_(keep_prob) + return x * random_tensor + +class DropPath(nn.Module): + """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks). + """ + def __init__(self, drop_prob: float = 0., scale_by_keep: bool = True): + super(DropPath, self).__init__() + self.drop_prob = drop_prob + self.scale_by_keep = scale_by_keep + + def forward(self, x): + return drop_path(x, self.drop_prob, self.training, self.scale_by_keep) + + def extra_repr(self): + return f'drop_prob={round(self.drop_prob,3):0.3f}' + +def _ntuple(n): + def parse(x): + from itertools import repeat + import collections.abc + if isinstance(x, collections.abc.Iterable) and not isinstance(x, str): + return tuple(x) + return tuple(repeat(x, n)) + return parse + +to_1tuple = _ntuple(1) +to_2tuple = _ntuple(2) +to_3tuple = _ntuple(3) +to_4tuple = _ntuple(4) +to_ntuple = _ntuple + +def _trunc_normal_(tensor, mean, std, a, b): + import warnings + import math + + def norm_cdf(x): + return (1. + math.erf(x / math.sqrt(2.))) / 2. + + if (mean < a - 2 * std) or (mean > b + 2 * std): + warnings.warn("mean is more than 2 std from [a, b] in nn.init.trunc_normal_. " + "The distribution of values may be incorrect.", + stacklevel=2) + + l = norm_cdf((a - mean) / std) + u = norm_cdf((b - mean) / std) + tensor.uniform_(2 * l - 1, 2 * u - 1) + tensor.erfinv_() + tensor.mul_(std * math.sqrt(2.)) + tensor.add_(mean) + tensor.clamp_(min=a, max=b) + return tensor + +def trunc_normal_(tensor, mean=0., std=1., a=-2., b=2.): + with torch.no_grad(): + return _trunc_normal_(tensor, mean, std, a, b) + +############# +# main swin # +############# + +class Mlp(nn.Module): + """ Multilayer perceptron.""" + + def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.): + super().__init__() + out_features = out_features or in_features + hidden_features = hidden_features or in_features + self.fc1 = nn.Linear(in_features, hidden_features) + self.act = act_layer() + self.fc2 = nn.Linear(hidden_features, out_features) + self.drop = nn.Dropout(drop) + + def forward(self, x): + x = self.fc1(x) + x = self.act(x) + x = self.drop(x) + x = self.fc2(x) + x = self.drop(x) + return x + + +def window_partition(x, window_size): + """ + Args: + x: (B, H, W, C) + window_size (int): window size + Returns: + windows: (num_windows*B, window_size, window_size, C) + """ + B, H, W, C = x.shape + x = x.view(B, H // window_size, window_size, W // window_size, window_size, C) + windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C) + return windows + + +def window_reverse(windows, window_size, H, W): + """ + Args: + windows: (num_windows*B, window_size, window_size, C) + window_size (int): Window size + H (int): Height of image + W (int): Width of image + Returns: + x: (B, H, W, C) + """ + B = int(windows.shape[0] / (H * W / window_size / window_size)) + x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1) + x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1) + return x + + +class WindowAttention(nn.Module): + """ Window based multi-head self attention (W-MSA) module with relative position bias. + It supports both of shifted and non-shifted window. + Args: + dim (int): Number of input channels. + window_size (tuple[int]): The height and width of the window. + num_heads (int): Number of attention heads. + qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True + qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set + attn_drop (float, optional): Dropout ratio of attention weight. Default: 0.0 + proj_drop (float, optional): Dropout ratio of output. Default: 0.0 + """ + + def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.): + + super().__init__() + self.dim = dim + self.window_size = window_size # Wh, Ww + self.num_heads = num_heads + head_dim = dim // num_heads + self.scale = qk_scale or head_dim ** -0.5 + + # define a parameter table of relative position bias + self.relative_position_bias_table = nn.Parameter( + torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads)) # 2*Wh-1 * 2*Ww-1, nH + + # get pair-wise relative position index for each token inside the window + coords_h = torch.arange(self.window_size[0]) + coords_w = torch.arange(self.window_size[1]) + coords = torch.stack(torch.meshgrid([coords_h, coords_w], indexing='ij')) # 2, Wh, Ww + coords_flatten = torch.flatten(coords, 1) # 2, Wh*Ww + relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # 2, Wh*Ww, Wh*Ww + relative_coords = relative_coords.permute(1, 2, 0).contiguous() # Wh*Ww, Wh*Ww, 2 + relative_coords[:, :, 0] += self.window_size[0] - 1 # shift to start from 0 + relative_coords[:, :, 1] += self.window_size[1] - 1 + relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1 + relative_position_index = relative_coords.sum(-1) # Wh*Ww, Wh*Ww + self.register_buffer("relative_position_index", relative_position_index) + + self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias) + self.attn_drop = nn.Dropout(attn_drop) + self.proj = nn.Linear(dim, dim) + self.proj_drop = nn.Dropout(proj_drop) + + trunc_normal_(self.relative_position_bias_table, std=.02) + self.softmax = nn.Softmax(dim=-1) + + def forward(self, x, mask=None): + """ Forward function. + Args: + x: input features with shape of (num_windows*B, N, C) + mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None + """ + B_, N, C = x.shape + qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4) + q, k, v = qkv[0], qkv[1], qkv[2] # make torchscript happy (cannot use tensor as tuple) + + q = q * self.scale + attn = (q @ k.transpose(-2, -1)) + + relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view( + self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1) # Wh*Ww,Wh*Ww,nH + relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous() # nH, Wh*Ww, Wh*Ww + attn = attn + relative_position_bias.unsqueeze(0) + + if mask is not None: + nW = mask.shape[0] + attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0) + attn = attn.view(-1, self.num_heads, N, N) + attn = self.softmax(attn) + else: + attn = self.softmax(attn) + + attn = self.attn_drop(attn) + + x = (attn @ v).transpose(1, 2).reshape(B_, N, C) + x = self.proj(x) + x = self.proj_drop(x) + return x + + +class SwinTransformerBlock(nn.Module): + """ Swin Transformer Block. + Args: + dim (int): Number of input channels. + num_heads (int): Number of attention heads. + window_size (int): Window size. + shift_size (int): Shift size for SW-MSA. + mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. + qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True + qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set. + drop (float, optional): Dropout rate. Default: 0.0 + attn_drop (float, optional): Attention dropout rate. Default: 0.0 + drop_path (float, optional): Stochastic depth rate. Default: 0.0 + act_layer (nn.Module, optional): Activation layer. Default: nn.GELU + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + """ + + def __init__(self, dim, num_heads, window_size=7, shift_size=0, + mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., drop_path=0., + act_layer=nn.GELU, norm_layer=nn.LayerNorm): + super().__init__() + self.dim = dim + self.num_heads = num_heads + self.window_size = window_size + self.shift_size = shift_size + self.mlp_ratio = mlp_ratio + assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size" + + self.norm1 = norm_layer(dim) + self.attn = WindowAttention( + dim, window_size=to_2tuple(self.window_size), num_heads=num_heads, + qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop) + + self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity() + self.norm2 = norm_layer(dim) + mlp_hidden_dim = int(dim * mlp_ratio) + self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop) + + self.H = None + self.W = None + + def forward(self, x, mask_matrix): + """ Forward function. + Args: + x: Input feature, tensor size (B, H*W, C). + H, W: Spatial resolution of the input feature. + mask_matrix: Attention mask for cyclic shift. + """ + B, L, C = x.shape + H, W = self.H, self.W + assert L == H * W, "input feature has wrong size" + + shortcut = x + x = self.norm1(x) + x = x.view(B, H, W, C) + + # pad feature maps to multiples of window size + pad_l = pad_t = 0 + pad_r = (self.window_size - W % self.window_size) % self.window_size + pad_b = (self.window_size - H % self.window_size) % self.window_size + x = F.pad(x, (0, 0, pad_l, pad_r, pad_t, pad_b)) + _, Hp, Wp, _ = x.shape + + # cyclic shift + if self.shift_size > 0: + shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2)) + attn_mask = mask_matrix + else: + shifted_x = x + attn_mask = None + + # partition windows + x_windows = window_partition(shifted_x, self.window_size) # nW*B, window_size, window_size, C + x_windows = x_windows.view(-1, self.window_size * self.window_size, C) # nW*B, window_size*window_size, C + + # W-MSA/SW-MSA + attn_windows = self.attn(x_windows, mask=attn_mask) # nW*B, window_size*window_size, C + + # merge windows + attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C) + shifted_x = window_reverse(attn_windows, self.window_size, Hp, Wp) # B H' W' C + + # reverse cyclic shift + if self.shift_size > 0: + x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2)) + else: + x = shifted_x + + if pad_r > 0 or pad_b > 0: + x = x[:, :H, :W, :].contiguous() + + x = x.view(B, H * W, C) + + # FFN + x = shortcut + self.drop_path(x) + x = x + self.drop_path(self.mlp(self.norm2(x))) + + return x + + +class PatchMerging(nn.Module): + """ Patch Merging Layer + Args: + dim (int): Number of input channels. + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + """ + def __init__(self, dim, norm_layer=nn.LayerNorm): + super().__init__() + self.dim = dim + self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False) + self.norm = norm_layer(4 * dim) + + def forward(self, x, H, W): + """ Forward function. + Args: + x: Input feature, tensor size (B, H*W, C). + H, W: Spatial resolution of the input feature. + """ + B, L, C = x.shape + assert L == H * W, "input feature has wrong size" + + x = x.view(B, H, W, C) + + # padding + pad_input = (H % 2 == 1) or (W % 2 == 1) + if pad_input: + x = F.pad(x, (0, 0, 0, W % 2, 0, H % 2)) + + x0 = x[:, 0::2, 0::2, :] # B H/2 W/2 C + x1 = x[:, 1::2, 0::2, :] # B H/2 W/2 C + x2 = x[:, 0::2, 1::2, :] # B H/2 W/2 C + x3 = x[:, 1::2, 1::2, :] # B H/2 W/2 C + x = torch.cat([x0, x1, x2, x3], -1) # B H/2 W/2 4*C + x = x.view(B, -1, 4 * C) # B H/2*W/2 4*C + + x = self.norm(x) + x = self.reduction(x) + + return x + + +class BasicLayer(nn.Module): + """ A basic Swin Transformer layer for one stage. + Args: + dim (int): Number of feature channels + depth (int): Depths of this stage. + num_heads (int): Number of attention head. + window_size (int): Local window size. Default: 7. + mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4. + qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True + qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set. + drop (float, optional): Dropout rate. Default: 0.0 + attn_drop (float, optional): Attention dropout rate. Default: 0.0 + drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0 + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None + use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False. + """ + + def __init__(self, + dim, + depth, + num_heads, + window_size=7, + mlp_ratio=4., + qkv_bias=True, + qk_scale=None, + drop=0., + attn_drop=0., + drop_path=0., + norm_layer=nn.LayerNorm, + downsample=None, + use_checkpoint=False): + super().__init__() + self.window_size = window_size + self.shift_size = window_size // 2 + self.depth = depth + self.use_checkpoint = use_checkpoint + + # build blocks + self.blocks = nn.ModuleList([ + SwinTransformerBlock( + dim=dim, + num_heads=num_heads, + window_size=window_size, + shift_size=0 if (i % 2 == 0) else window_size // 2, + mlp_ratio=mlp_ratio, + qkv_bias=qkv_bias, + qk_scale=qk_scale, + drop=drop, + attn_drop=attn_drop, + drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path, + norm_layer=norm_layer) + for i in range(depth)]) + + # patch merging layer + if downsample is not None: + self.downsample = downsample(dim=dim, norm_layer=norm_layer) + else: + self.downsample = None + + def forward(self, x, H, W): + """ Forward function. + Args: + x: Input feature, tensor size (B, H*W, C). + H, W: Spatial resolution of the input feature. + """ + + # calculate attention mask for SW-MSA + Hp = int(np.ceil(H / self.window_size)) * self.window_size + Wp = int(np.ceil(W / self.window_size)) * self.window_size + img_mask = torch.zeros((1, Hp, Wp, 1), device=x.device, dtype=x.dtype) # 1 Hp Wp 1 + h_slices = (slice(0, -self.window_size), + slice(-self.window_size, -self.shift_size), + slice(-self.shift_size, None)) + w_slices = (slice(0, -self.window_size), + slice(-self.window_size, -self.shift_size), + slice(-self.shift_size, None)) + cnt = 0 + for h in h_slices: + for w in w_slices: + img_mask[:, h, w, :] = cnt + cnt += 1 + + mask_windows = window_partition(img_mask, self.window_size) # nW, window_size, window_size, 1 + mask_windows = mask_windows.view(-1, self.window_size * self.window_size) + attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2) + attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0)) + + for blk in self.blocks: + blk.H, blk.W = H, W + if self.use_checkpoint: + x = checkpoint.checkpoint(blk, x, attn_mask) + else: + x = blk(x, attn_mask) + if self.downsample is not None: + x_down = self.downsample(x, H, W) + Wh, Ww = (H + 1) // 2, (W + 1) // 2 + return x, H, W, x_down, Wh, Ww + else: + return x, H, W, x, H, W + + +class PatchEmbed(nn.Module): + """ Image to Patch Embedding + Args: + patch_size (int): Patch token size. Default: 4. + in_chans (int): Number of input image channels. Default: 3. + embed_dim (int): Number of linear projection output channels. Default: 96. + norm_layer (nn.Module, optional): Normalization layer. Default: None + """ + + def __init__(self, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None): + super().__init__() + patch_size = to_2tuple(patch_size) + self.patch_size = patch_size + + self.in_chans = in_chans + self.embed_dim = embed_dim + + self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size) + if norm_layer is not None: + self.norm = norm_layer(embed_dim) + else: + self.norm = None + + def forward(self, x): + """Forward function.""" + # padding + _, _, H, W = x.size() + if W % self.patch_size[1] != 0: + x = F.pad(x, (0, self.patch_size[1] - W % self.patch_size[1])) + if H % self.patch_size[0] != 0: + x = F.pad(x, (0, 0, 0, self.patch_size[0] - H % self.patch_size[0])) + + x = self.proj(x) # B C Wh Ww + if self.norm is not None: + Wh, Ww = x.size(2), x.size(3) + x = x.flatten(2).transpose(1, 2) + x = self.norm(x) + x = x.transpose(1, 2).view(-1, self.embed_dim, Wh, Ww) + + return x + + +class SwinTransformer(nn.Module): + """ Swin Transformer backbone. + A PyTorch impl of : `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows` - + https://arxiv.org/pdf/2103.14030 + Args: + pretrain_img_size (int): Input image size for training the pretrained model, + used in absolute postion embedding. Default 224. + patch_size (int | tuple(int)): Patch size. Default: 4. + in_chans (int): Number of input image channels. Default: 3. + embed_dim (int): Number of linear projection output channels. Default: 96. + depths (tuple[int]): Depths of each Swin Transformer stage. + num_heads (tuple[int]): Number of attention head of each stage. + window_size (int): Window size. Default: 7. + mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4. + qkv_bias (bool): If True, add a learnable bias to query, key, value. Default: True + qk_scale (float): Override default qk scale of head_dim ** -0.5 if set. + drop_rate (float): Dropout rate. + attn_drop_rate (float): Attention dropout rate. Default: 0. + drop_path_rate (float): Stochastic depth rate. Default: 0.2. + norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm. + ape (bool): If True, add absolute position embedding to the patch embedding. Default: False. + patch_norm (bool): If True, add normalization after patch embedding. Default: True. + out_indices (Sequence[int]): Output from which stages. + frozen_stages (int): Stages to be frozen (stop grad and set eval mode). + -1 means not freezing any parameters. + use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False. + """ + + def __init__(self, + pretrain_img_size=224, + patch_size=4, + in_chans=3, + embed_dim=96, + depths=[2, 2, 6, 2], + num_heads=[3, 6, 12, 24], + window_size=7, + mlp_ratio=4., + qkv_bias=True, + qk_scale=None, + drop_rate=0., + attn_drop_rate=0., + drop_path_rate=0.2, + norm_layer=nn.LayerNorm, + ape=False, + patch_norm=True, + out_indices=(0, 1, 2, 3), + frozen_stages=-1, + use_checkpoint=False): + super().__init__() + + self.pretrain_img_size = pretrain_img_size + self.num_layers = len(depths) + self.embed_dim = embed_dim + self.ape = ape + self.patch_norm = patch_norm + self.out_indices = out_indices + self.frozen_stages = frozen_stages + + # split image into non-overlapping patches + self.patch_embed = PatchEmbed( + patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim, + norm_layer=norm_layer if self.patch_norm else None) + + # absolute position embedding + if self.ape: + pretrain_img_size = to_2tuple(pretrain_img_size) + patch_size = to_2tuple(patch_size) + patches_resolution = [pretrain_img_size[0] // patch_size[0], pretrain_img_size[1] // patch_size[1]] + + self.absolute_pos_embed = nn.Parameter(torch.zeros(1, embed_dim, patches_resolution[0], patches_resolution[1])) + trunc_normal_(self.absolute_pos_embed, std=.02) + + self.pos_drop = nn.Dropout(p=drop_rate) + + # stochastic depth + dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))] # stochastic depth decay rule + + # build layers + self.layers = nn.ModuleList() + for i_layer in range(self.num_layers): + layer = BasicLayer( + dim=int(embed_dim * 2 ** i_layer), + depth=depths[i_layer], + num_heads=num_heads[i_layer], + window_size=window_size, + mlp_ratio=mlp_ratio, + qkv_bias=qkv_bias, + qk_scale=qk_scale, + drop=drop_rate, + attn_drop=attn_drop_rate, + drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])], + norm_layer=norm_layer, + downsample=PatchMerging if (i_layer < self.num_layers - 1) else None, + use_checkpoint=use_checkpoint) + self.layers.append(layer) + + num_features = [int(embed_dim * 2 ** i) for i in range(self.num_layers)] + self.num_features = num_features + + # add a norm layer for each output + for i_layer in out_indices: + layer = norm_layer(num_features[i_layer]) + layer_name = f'norm{i_layer}' + self.add_module(layer_name, layer) + + self._freeze_stages() + + def _freeze_stages(self): + if self.frozen_stages >= 0: + self.patch_embed.eval() + for param in self.patch_embed.parameters(): + param.requires_grad = False + + if self.frozen_stages >= 1 and self.ape: + self.absolute_pos_embed.requires_grad = False + + if self.frozen_stages >= 2: + self.pos_drop.eval() + for i in range(0, self.frozen_stages - 1): + m = self.layers[i] + m.eval() + for param in m.parameters(): + param.requires_grad = False + + def forward(self, x): + """Forward function.""" + x = self.patch_embed(x) + + Wh, Ww = x.size(2), x.size(3) + if self.ape: + # interpolate the position embedding to the corresponding size + absolute_pos_embed = F.interpolate(self.absolute_pos_embed, size=(Wh, Ww), mode='bicubic') + x = (x + absolute_pos_embed).flatten(2).transpose(1, 2) # B Wh*Ww C + else: + x = x.flatten(2).transpose(1, 2) + x = self.pos_drop(x) + + outs = [] + for i in range(self.num_layers): + layer = self.layers[i] + x_out, H, W, x, Wh, Ww = layer(x, Wh, Ww) + + if i in self.out_indices: + norm_layer = getattr(self, f'norm{i}') + x_out = norm_layer(x_out) + + out = x_out.view(-1, H, W, self.num_features[i]).permute(0, 3, 1, 2).contiguous() + outs.append(out) + + outputs = { + 'res2' : outs[0], + 'res3' : outs[1], + 'res4' : outs[2], + 'res5' : outs[3],} + return outputs + + def train(self, mode=True): + """Convert the model into training mode while keep layers freezed.""" + super(SwinTransformer, self).train(mode) + self._freeze_stages() + return self diff --git a/nodes/load_seecoder_node.py b/nodes/load_seecoder_node.py new file mode 100644 index 0000000..4a19f73 --- /dev/null +++ b/nodes/load_seecoder_node.py @@ -0,0 +1,36 @@ +import os + +from safetensors.torch import load_file + +from folder_paths import models_dir + +from ..modules.seecoder import SemanticContextEncoder + + +SEECODER_PATH = os.path.join(models_dir, 'seecoder') +os.makedirs(SEECODER_PATH, exist_ok=True) + + +class LoadSeeCoderNode: + @classmethod + def INPUT_TYPES(s): + return {"required": { "checkpoint": (os.listdir(SEECODER_PATH), )}} + + RETURN_TYPES = ("SEECODER",) + FUNCTION = "load" + + CATEGORY = "seecoder" + + def load(self, checkpoint): + checkpoint_path = os.path.join(SEECODER_PATH, checkpoint) + state_dict = load_file(checkpoint_path) + cleaned_state_dict = {} + has_pe = False + for key in state_dict.keys(): + cleaned_key = key.replace('ctx.image.', '') + if 'pe_layer' in cleaned_key: + has_pe = True + cleaned_state_dict[cleaned_key] = state_dict[key] + seecoder = SemanticContextEncoder(has_pe) + seecoder.load_state_dict(cleaned_state_dict) + return (seecoder,) \ No newline at end of file diff --git a/nodes/load_seecoder_uncond_node.py b/nodes/load_seecoder_uncond_node.py new file mode 100644 index 0000000..8f2594d --- /dev/null +++ b/nodes/load_seecoder_uncond_node.py @@ -0,0 +1,25 @@ +import os + +import torch + +from folder_paths import models_dir + + +SEECODER_PATH = os.path.join(models_dir, 'seecoder') +os.makedirs(SEECODER_PATH, exist_ok=True) + + +class LoadSeeCoderUncondNode: + @classmethod + def INPUT_TYPES(s): + return {"required": { "pth": (os.listdir(SEECODER_PATH), )}} + + RETURN_TYPES = ("CONDITIONING",) + FUNCTION = "load" + + CATEGORY = "seecoder" + + def load(self, pth): + checkpoint_path = os.path.join(SEECODER_PATH, pth) + conditioning = torch.load(checkpoint_path).unsqueeze(0) + return ([[conditioning, {}]],) \ No newline at end of file diff --git a/nodes/seecoder_encode_node.py b/nodes/seecoder_encode_node.py new file mode 100644 index 0000000..c2382be --- /dev/null +++ b/nodes/seecoder_encode_node.py @@ -0,0 +1,29 @@ +import comfy.model_management + + +class SeecoderEncodeNode: + @classmethod + def INPUT_TYPES(s): + return { + "required": { + "seecoder": ("SEECODER",), + "image": ("IMAGE",), + }, + } + RETURN_TYPES = ("CONDITIONING",) + FUNCTION = "encode" + CATEGORY = "seecoder" + + def encode(self, + seecoder, + image): + + intermediate_device = comfy.model_management.intermediate_device() + torch_device = comfy.model_management.get_torch_device() + + seecoder.to(torch_device) + # latent = latent_image['samples'].to(torch_device) + embed = seecoder(image.permute(0,3,1,2).to(torch_device)).to(intermediate_device) + seecoder.to(intermediate_device) + + return ([[embed, {}]], ) \ No newline at end of file