# --------------------------------------------------------------- # Copyright (c) 2021, NVIDIA Corporation. All rights reserved. # # This work is licensed under the NVIDIA Source Code License # --------------------------------------------------------------- import torch import torch.distributed as dist import torch.nn as nn import torch.nn.functional as F from functools import partial import warnings from einops import rearrange from timm.models.layers import DropPath, to_2tuple, trunc_normal_ from timm.models.registry import register_model from timm.models.vision_transformer import _cfg import math from mmcv.cnn import ConvModule from utils.commons.hparams import hparams def resize(input, size=None, scale_factor=None, mode='nearest', align_corners=None, warning=True): if warning: if size is not None and align_corners: input_h, input_w = tuple(int(x) for x in input.shape[2:]) output_h, output_w = tuple(int(x) for x in size) if output_h > input_h or output_w > output_h: if ((output_h > 1 and output_w > 1 and input_h > 1 and input_w > 1) and (output_h - 1) % (input_h - 1) and (output_w - 1) % (input_w - 1)): warnings.warn( f'When align_corners={align_corners}, ' 'the output would more aligned if ' f'input size {(input_h, input_w)} is `x+1` and ' f'out size {(output_h, output_w)} is `nx+1`') if isinstance(size, torch.Size): size = tuple(int(x) for x in size) return F.interpolate(input, size, scale_factor, mode, align_corners) class HeadMLP(nn.Module): """ Linear Embedding """ def __init__(self, input_dim=2048, embed_dim=768): super().__init__() self.proj = nn.Linear(input_dim, embed_dim) def forward(self, x): x = x.flatten(2).transpose(1, 2) x = self.proj(x) return x class Mlp(nn.Module): 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.dwconv = DWConv(hidden_features) self.act = act_layer() self.fc2 = nn.Linear(hidden_features, out_features) self.drop = nn.Dropout(drop) self.apply(self._init_weights) def _init_weights(self, m): if isinstance(m, nn.Linear): trunc_normal_(m.weight, std=.02) if isinstance(m, nn.Linear) and m.bias is not None: nn.init.constant_(m.bias, 0) elif isinstance(m, nn.LayerNorm): nn.init.constant_(m.bias, 0) nn.init.constant_(m.weight, 1.0) elif isinstance(m, nn.Conv2d): fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels fan_out //= m.groups m.weight.data.normal_(0, math.sqrt(2.0 / fan_out)) if m.bias is not None: m.bias.data.zero_() def forward(self, x, H, W): x = self.fc1(x) x = self.dwconv(x, H, W) x = self.act(x) x = self.drop(x) x = self.fc2(x) x = self.drop(x) return x class Attention(nn.Module): def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0., sr_ratio=1): super().__init__() assert dim % num_heads == 0, f"dim {dim} should be divided by num_heads {num_heads}." self.dim = dim self.num_heads = num_heads head_dim = dim // num_heads self.scale = qk_scale or head_dim ** -0.5 self.q = nn.Linear(dim, dim, bias=qkv_bias) self.kv = nn.Linear(dim, dim * 2, bias=qkv_bias) self.attn_drop = nn.Dropout(attn_drop) self.proj = nn.Linear(dim, dim) self.proj_drop = nn.Dropout(proj_drop) self.sr_ratio = sr_ratio if sr_ratio > 1: self.sr = nn.Conv2d(dim, dim, kernel_size=sr_ratio, stride=sr_ratio) self.norm = nn.LayerNorm(dim) self.apply(self._init_weights) def _init_weights(self, m): if isinstance(m, nn.Linear): trunc_normal_(m.weight, std=.02) if isinstance(m, nn.Linear) and m.bias is not None: nn.init.constant_(m.bias, 0) elif isinstance(m, nn.LayerNorm): nn.init.constant_(m.bias, 0) nn.init.constant_(m.weight, 1.0) elif isinstance(m, nn.Conv2d): fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels fan_out //= m.groups m.weight.data.normal_(0, math.sqrt(2.0 / fan_out)) if m.bias is not None: m.bias.data.zero_() def forward(self, x, H, W): B, N, C = x.shape q = self.q(x).reshape(B, N, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3) if self.sr_ratio > 1: x_ = x.permute(0, 2, 1).reshape(B, C, H, W) x_ = self.sr(x_).reshape(B, C, -1).permute(0, 2, 1) x_ = self.norm(x_) kv = self.kv(x_).reshape(B, -1, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4) else: kv = self.kv(x).reshape(B, -1, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4) k, v = kv[0], kv[1] attn = (q @ k.transpose(-2, -1)) * self.scale attn = attn.softmax(dim=-1) 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 Block(nn.Module): def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0., drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm, sr_ratio=1): super().__init__() self.norm1 = norm_layer(dim) self.attn = Attention( dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop, sr_ratio=sr_ratio) # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here 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.apply(self._init_weights) def _init_weights(self, m): if isinstance(m, nn.Linear): trunc_normal_(m.weight, std=.02) if isinstance(m, nn.Linear) and m.bias is not None: nn.init.constant_(m.bias, 0) elif isinstance(m, nn.LayerNorm): nn.init.constant_(m.bias, 0) nn.init.constant_(m.weight, 1.0) elif isinstance(m, nn.Conv2d): fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels fan_out //= m.groups m.weight.data.normal_(0, math.sqrt(2.0 / fan_out)) if m.bias is not None: m.bias.data.zero_() def forward(self, x, H, W): x = x + self.drop_path(self.attn(self.norm1(x), H, W)) x = x + self.drop_path(self.mlp(self.norm2(x), H, W)) return x class OverlapPatchEmbed(nn.Module): """ Image to Patch Embedding """ def __init__(self, img_size=224, patch_size=7, stride=4, in_chans=3, embed_dim=768): super().__init__() img_size = to_2tuple(img_size) patch_size = to_2tuple(patch_size) self.img_size = img_size self.patch_size = patch_size self.H, self.W = img_size[0] // patch_size[0], img_size[1] // patch_size[1] self.num_patches = self.H * self.W self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=stride, padding=(patch_size[0] // 2, patch_size[1] // 2)) self.norm = nn.LayerNorm(embed_dim) self.apply(self._init_weights) def _init_weights(self, m): if isinstance(m, nn.Linear): trunc_normal_(m.weight, std=.02) if isinstance(m, nn.Linear) and m.bias is not None: nn.init.constant_(m.bias, 0) elif isinstance(m, nn.LayerNorm): nn.init.constant_(m.bias, 0) nn.init.constant_(m.weight, 1.0) elif isinstance(m, nn.Conv2d): fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels fan_out //= m.groups m.weight.data.normal_(0, math.sqrt(2.0 / fan_out)) if m.bias is not None: m.bias.data.zero_() def forward(self, x): x = self.proj(x) _, _, H, W = x.shape x = x.flatten(2).transpose(1, 2) x = self.norm(x) return x, H, W class MixVisionTransformer(nn.Module): def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dims=[64, 128, 256, 512], num_heads=[1, 2, 4, 8], mlp_ratios=[4, 4, 4, 4], qkv_bias=False, qk_scale=None, drop_rate=0., attn_drop_rate=0., drop_path_rate=0., norm_layer=nn.LayerNorm, depths=[3, 4, 6, 3], sr_ratios=[8, 4, 2, 1]): super().__init__() self.num_classes = num_classes self.depths = depths # patch_embed self.patch_embed1 = OverlapPatchEmbed(img_size=img_size, patch_size=7, stride=4, in_chans=in_chans, embed_dim=embed_dims[0]) self.patch_embed2 = OverlapPatchEmbed(img_size=img_size // 4, patch_size=3, stride=2, in_chans=embed_dims[0], embed_dim=embed_dims[1]) self.patch_embed3 = OverlapPatchEmbed(img_size=img_size // 8, patch_size=3, stride=2, in_chans=embed_dims[1], embed_dim=embed_dims[2]) self.patch_embed4 = OverlapPatchEmbed(img_size=img_size // 16, patch_size=3, stride=2, in_chans=embed_dims[2], embed_dim=embed_dims[3]) # transformer encoder dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))] # stochastic depth decay rule cur = 0 self.block1 = nn.ModuleList([Block( dim=embed_dims[0], num_heads=num_heads[0], mlp_ratio=mlp_ratios[0], qkv_bias=qkv_bias, qk_scale=qk_scale, drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer, sr_ratio=sr_ratios[0]) for i in range(depths[0])]) self.norm1 = norm_layer(embed_dims[0]) cur += depths[0] self.block2 = nn.ModuleList([Block( dim=embed_dims[1], num_heads=num_heads[1], mlp_ratio=mlp_ratios[1], qkv_bias=qkv_bias, qk_scale=qk_scale, drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer, sr_ratio=sr_ratios[1]) for i in range(depths[1])]) self.norm2 = norm_layer(embed_dims[1]) cur += depths[1] self.block3 = nn.ModuleList([Block( dim=embed_dims[2], num_heads=num_heads[2], mlp_ratio=mlp_ratios[2], qkv_bias=qkv_bias, qk_scale=qk_scale, drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer, sr_ratio=sr_ratios[2]) for i in range(depths[2])]) self.norm3 = norm_layer(embed_dims[2]) cur += depths[2] self.block4 = nn.ModuleList([Block( dim=embed_dims[3], num_heads=num_heads[3], mlp_ratio=mlp_ratios[3], qkv_bias=qkv_bias, qk_scale=qk_scale, drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer, sr_ratio=sr_ratios[3]) for i in range(depths[3])]) self.norm4 = norm_layer(embed_dims[3]) # classification head # self.head = nn.Linear(embed_dims[3], num_classes) if num_classes > 0 else nn.Identity() self.apply(self._init_weights) def _init_weights(self, m): if isinstance(m, nn.Linear): trunc_normal_(m.weight, std=.02) if isinstance(m, nn.Linear) and m.bias is not None: nn.init.constant_(m.bias, 0) elif isinstance(m, nn.LayerNorm): nn.init.constant_(m.bias, 0) nn.init.constant_(m.weight, 1.0) elif isinstance(m, nn.Conv2d): fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels fan_out //= m.groups m.weight.data.normal_(0, math.sqrt(2.0 / fan_out)) if m.bias is not None: m.bias.data.zero_() def reset_drop_path(self, drop_path_rate): dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(self.depths))] cur = 0 for i in range(self.depths[0]): self.block1[i].drop_path.drop_prob = dpr[cur + i] cur += self.depths[0] for i in range(self.depths[1]): self.block2[i].drop_path.drop_prob = dpr[cur + i] cur += self.depths[1] for i in range(self.depths[2]): self.block3[i].drop_path.drop_prob = dpr[cur + i] cur += self.depths[2] for i in range(self.depths[3]): self.block4[i].drop_path.drop_prob = dpr[cur + i] def freeze_patch_emb(self): self.patch_embed1.requires_grad = False @torch.jit.ignore def no_weight_decay(self): return {'pos_embed1', 'pos_embed2', 'pos_embed3', 'pos_embed4', 'cls_token'} # has pos_embed may be better def get_classifier(self): return self.head def reset_classifier(self, num_classes, global_pool=''): self.num_classes = num_classes self.head = nn.Linear(self.embed_dim, num_classes) if num_classes > 0 else nn.Identity() def forward_features(self, x): B = x.shape[0] outs = [] # stage 1 x, H, W = self.patch_embed1(x) for i, blk in enumerate(self.block1): x = blk(x, H, W) x = self.norm1(x) x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous() outs.append(x) # stage 2 x, H, W = self.patch_embed2(x) for i, blk in enumerate(self.block2): x = blk(x, H, W) x = self.norm2(x) x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous() outs.append(x) # stage 3 x, H, W = self.patch_embed3(x) for i, blk in enumerate(self.block3): x = blk(x, H, W) x = self.norm3(x) x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous() outs.append(x) # stage 4 x, H, W = self.patch_embed4(x) for i, blk in enumerate(self.block4): x = blk(x, H, W) x = self.norm4(x) x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous() outs.append(x) return outs def forward(self, x): x = self.forward_features(x) # x = self.head(x) return x class DWConv(nn.Module): def __init__(self, dim=768): super(DWConv, self).__init__() self.dwconv = nn.Conv2d(dim, dim, 3, 1, 1, bias=True, groups=dim) def forward(self, x, H, W): B, N, C = x.shape x = x.transpose(1, 2).view(B, C, H, W) x = self.dwconv(x) x = x.flatten(2).transpose(1, 2) return x class mit_b0(MixVisionTransformer): # 3.319M def __init__(self, **kwargs): super(mit_b0, self).__init__( patch_size=4, embed_dims=[32, 64, 160, 256], num_heads=[1, 2, 5, 8], mlp_ratios=[4, 4, 4, 4], qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[2, 2, 2, 2], sr_ratios=[8, 4, 2, 1], drop_rate=0.0, drop_path_rate=0.1) self.load_state_dict(torch.load('checkpoints/pretrained_ckpts/mit_b0.pth'), strict=False) class mit_b1(MixVisionTransformer): # 13.151M def __init__(self, **kwargs): super(mit_b1, self).__init__( patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[4, 4, 4, 4], qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[2, 2, 2, 2], sr_ratios=[8, 4, 2, 1], drop_rate=0.0, drop_path_rate=0.1) self.load_state_dict(torch.load('checkpoints/pretrained_ckpts/mit_b1.pth'), strict=False) class mit_b2(MixVisionTransformer): # 24.196M def __init__(self, **kwargs): super(mit_b2, self).__init__( patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[4, 4, 4, 4], qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[3, 4, 6, 3], sr_ratios=[8, 4, 2, 1], drop_rate=0.0, drop_path_rate=0.1) self.load_state_dict(torch.load('checkpoints/pretrained_ckpts/mit_b2.pth'), strict=False) class mit_b3(MixVisionTransformer): # 44.072M def __init__(self, **kwargs): super(mit_b3, self).__init__( patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[4, 4, 4, 4], qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[3, 4, 18, 3], sr_ratios=[8, 4, 2, 1], drop_rate=0.0, drop_path_rate=0.1) self.load_state_dict(torch.load('checkpoints/pretrained_ckpts/mit_b3.pth'), strict=False) class mit_b4(MixVisionTransformer): # 60.843M def __init__(self, **kwargs): super(mit_b4, self).__init__( patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[4, 4, 4, 4], qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[3, 8, 27, 3], sr_ratios=[8, 4, 2, 1], drop_rate=0.0, drop_path_rate=0.1) self.load_state_dict(torch.load('checkpoints/pretrained_ckpts/mit_b4.pth'), strict=False) class mit_b5(MixVisionTransformer): # 81.443M def __init__(self, **kwargs): super(mit_b5, self).__init__( patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[4, 4, 4, 4], qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[3, 6, 40, 3], sr_ratios=[8, 4, 2, 1], drop_rate=0.0, drop_path_rate=0.1) self.load_state_dict(torch.load('checkpoints/pretrained_ckpts/mit_b5.pth'), strict=False) class SegFormerHead(nn.Module): """ SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers """ def __init__(self, segformer_scale='b3'): super().__init__() self.segformer_scale = segformer_scale self.in_channels = [64, 128, 320, 512] if self.segformer_scale != 'b0' else [32, 64, 160, 256] self.feature_strides = [4, 8, 16, 32] self.in_index = [0, 1, 2, 3] self.input_transform='multiple_select' self.dropout = nn.Dropout2d(0.1) c1_in_channels, c2_in_channels, c3_in_channels, c4_in_channels = self.in_channels embedding_dim = self.embedding_dim = 256 self.linear_c4 = HeadMLP(input_dim=c4_in_channels, embed_dim=embedding_dim) self.linear_c3 = HeadMLP(input_dim=c3_in_channels, embed_dim=embedding_dim) self.linear_c2 = HeadMLP(input_dim=c2_in_channels, embed_dim=embedding_dim) self.linear_c1 = HeadMLP(input_dim=c1_in_channels, embed_dim=embedding_dim) if dist.is_initialized(): self.linear_fuse = ConvModule( in_channels=embedding_dim*4, out_channels=embedding_dim, kernel_size=1, norm_cfg=dict(type='SyncBN', requires_grad=True) ) else: self.linear_fuse = ConvModule( in_channels=embedding_dim*4, out_channels=embedding_dim, kernel_size=1, norm_cfg=dict(type='BN', requires_grad=True) ) def _transform_inputs(self, inputs): """Transform inputs for decoder. Args: inputs (list[Tensor]): List of multi-level img features. Returns: Tensor: The transformed inputs """ if self.input_transform == 'multiple_select': inputs = [inputs[i] for i in self.in_index] else: inputs = inputs[self.in_index] return inputs def forward(self, inputs): x = self._transform_inputs(inputs) # len=4, 1/4,1/8,1/16,1/32 c1, c2, c3, c4 = x ############## MLP decoder on C1-C4 ########### n, _, h, w = c4.shape _c4 = self.linear_c4(c4).permute(0,2,1).reshape(n, -1, c4.shape[2], c4.shape[3]) _c4 = resize(_c4, size=c1.size()[2:],mode='bilinear',align_corners=False) _c3 = self.linear_c3(c3).permute(0,2,1).reshape(n, -1, c3.shape[2], c3.shape[3]) _c3 = resize(_c3, size=c1.size()[2:],mode='bilinear',align_corners=False) _c2 = self.linear_c2(c2).permute(0,2,1).reshape(n, -1, c2.shape[2], c2.shape[3]) _c2 = resize(_c2, size=c1.size()[2:],mode='bilinear',align_corners=False) _c1 = self.linear_c1(c1).permute(0,2,1).reshape(n, -1, c1.shape[2], c1.shape[3]) _c = self.linear_fuse(torch.cat([_c4, _c3, _c2, _c1], dim=1)) x = self.dropout(_c) return x # from modules.hidenerf.models.networks_stylegan2 import Conv2dLayer from modules.eg3ds.models.networks_stylegan2 import Conv2dLayer class conv(nn.Module): def __init__(self, num_in_layers, num_out_layers, kernel_size, up=1, down=1): super(conv, self).__init__() self.conv = Conv2dLayer(num_in_layers, num_out_layers, kernel_size, activation='elu', up=up, down=down) self.bn = nn.InstanceNorm2d( num_out_layers, track_running_stats=False, affine=True ) def forward(self, x): return self.bn(self.conv(x)) class SegFormerImg2PlaneBackbone(nn.Module): def __init__(self, mode='b3'): super().__init__() mode2cls = { 'b0': mit_b0, 'b1': mit_b1, 'b2': mit_b2, 'b3': mit_b3, 'b4': mit_b4, 'b5': mit_b5, } self.mode = mode self.mix_vit = mode2cls[mode]() self.fuse_head = SegFormerHead(mode) self.to_plane_cnn = nn.Sequential(*[ nn.Conv2d(in_channels=256, out_channels=256, kernel_size=3, stride=1, padding=1), nn.LeakyReLU(negative_slope=0.01, inplace=True), nn.Conv2d(in_channels=256, out_channels=256, kernel_size=3, stride=1, padding=1), nn.LeakyReLU(negative_slope=0.01, inplace=True), nn.Conv2d(in_channels=256, out_channels=256, kernel_size=3, stride=1, padding=1), nn.LeakyReLU(negative_slope=0.01, inplace=True), nn.UpsamplingBilinear2d(scale_factor=2.), nn.Conv2d(in_channels=256, out_channels=96, kernel_size=3, stride=1, padding=1), ]) def forward(self, x): """ x: [B, 3, H=512, W=512] return: plane: [B, 96, H=256, W=256] """ feats = self.mix_vit(x) fused_feat = self.fuse_head(feats) planes = self.to_plane_cnn(fused_feat) planes = planes.view(len(planes), 3, -1, planes.shape[-2], planes.shape[-1]) planes_xy = planes[:,0] planes_xy = torch.flip(planes_xy, [2]) planes_xz = planes[:,1] planes_xz = torch.flip(planes_xz, [2]) planes_zy = planes[:,2] planes_zy = torch.flip(planes_zy, [2, 3]) planes = torch.stack([planes_xy, planes_xz, planes_zy], dim=1) # [N, 3, C, H, W] return planes class TemporalAttNet(nn.Module): """ Used to smooth the secc_plane with a window input """ def __init__(self, in_dim=96, seq_len=5): super().__init__() self.seq_len = seq_len self.conv2d_layers = nn.Sequential(*[ # [B, C=96, T, H=224, W=224] ==> [B, 64, T, 112, 112] nn.Conv3d(in_dim, 64, kernel_size=(1, 3, 3), stride=1, padding=(0, 1, 1)), nn.LeakyReLU(0.02, True), nn.Conv3d(64, 64, kernel_size=(1, 3, 3), stride=1, padding=(0, 1, 1)), nn.LeakyReLU(0.02, True), nn.AvgPool3d(kernel_size=(1, 2, 2), stride=(1,2,2), count_include_pad=False), # [B, C=64, T, H=112, W=112] ==> [B, 32, T, 56, 56] nn.Conv3d(64, 32, kernel_size=(1, 3, 3), stride=1, padding=(0, 1, 1)), nn.LeakyReLU(0.02, True), nn.Conv3d(32, 32, kernel_size=(1, 3, 3), stride=1, padding=(0, 1, 1)), nn.LeakyReLU(0.02, True), nn.AvgPool3d(kernel_size=(1, 2, 2), stride=(1,2,2), count_include_pad=False), # [B, C=32, T, H=56, W=56] ==> [B, 16, T, 28, 28] nn.Conv3d(32, 16, kernel_size=(1, 3, 3), stride=1, padding=(0, 1, 1)), nn.LeakyReLU(0.02, True), nn.Conv3d(16, 16, kernel_size=(1, 3, 3), stride=1, padding=(0, 1, 1)), nn.LeakyReLU(0.02, True), nn.AvgPool3d(kernel_size=(1, 2, 2), stride=(1,2,2), count_include_pad=False), ]) self.conv3d_layers = nn.Sequential(*[ # [B, C=16, T, H=28, W=28] ==> [B, 8, T, 14, 14] nn.Conv3d(16, 8, kernel_size=3, stride=1, padding=1), nn.LeakyReLU(0.02, True), nn.AvgPool3d(kernel_size=(1, 2, 2), stride=(1,2,2), count_include_pad=False), # [B, C=8, T, H=14, W=14] ==> [B, 8, T, 7, 7] nn.Conv3d(8, 8, kernel_size=3, stride=1, padding=1), nn.LeakyReLU(0.02, True), nn.AvgPool3d(kernel_size=(1, 2, 2), stride=(1,2,2), count_include_pad=False), # [B, C=8, T, H=7, W=7] ==> [B, 4, T, 1, 1] nn.Conv3d(8, 4, kernel_size=3, stride=1, padding=1), nn.LeakyReLU(0.02, True), nn.Conv3d(4, 2, kernel_size=3, stride=1, padding=1), nn.LeakyReLU(0.02, True), nn.Conv3d(2, 1, kernel_size=3, stride=1, padding=1), nn.LeakyReLU(0.02, True), nn.AvgPool3d(kernel_size=(1, 7, 7), stride=1, count_include_pad=False), ]) self.to_attention_weights = nn.Sequential( nn.Linear(in_features=self.seq_len, out_features=self.seq_len, bias=True), nn.Softmax(dim=1) ) def forward(self, x): """ x: [B, C, T, H, W] y: [B, T] attention weights out: [B, C, H, W] """ b,c,t,h,w = x.shape y = F.interpolate(x, size=(t, 224, 224), mode='trilinear') y = self.conv2d_layers(y) # [B, 16, 5, 28, 28] y = self.conv3d_layers(y) # [B, 1, T, 1, 1] y = y.squeeze(1, 3, 4) # [B, T] assert y.ndim == 2 y = y.reshape([b, 1, t, 1, 1]) out = (y * x).sum(dim=2) return out class SegFormerSECC2PlaneBackbone(nn.Module): def __init__(self, mode='b0', out_channels=96, pncc_cond_mode='cano_src_tgt'): super().__init__() mode2cls = { 'b0': mit_b0, 'b1': mit_b1, 'b2': mit_b2, 'b3': mit_b3, 'b4': mit_b4, 'b5': mit_b5, } self.mode = mode self.pncc_cond_mode = pncc_cond_mode in_dim = 9 if pncc_cond_mode == 'cano_src_tgt' else 6 self.prenet = Conv2dLayer(in_dim, 3, 1) self.mix_vit = mode2cls[mode]() self.fuse_head = SegFormerHead(mode) self.to_plane_cnn = nn.Sequential(*[ nn.Conv2d(in_channels=256, out_channels=256, kernel_size=3, stride=1, padding=1), nn.LeakyReLU(negative_slope=0.01, inplace=True), nn.Conv2d(in_channels=256, out_channels=256, kernel_size=3, stride=1, padding=1), nn.LeakyReLU(negative_slope=0.01, inplace=True), nn.Conv2d(in_channels=256, out_channels=256, kernel_size=3, stride=1, padding=1), nn.LeakyReLU(negative_slope=0.01, inplace=True), nn.UpsamplingBilinear2d(scale_factor=2.), nn.Conv2d(in_channels=256, out_channels=out_channels, kernel_size=3, stride=1, padding=1), ]) # if hparams['use_motion_smo_net']: # self.motion_smo_win_size = hparams['motion_smo_win_size'] # self.smo_net = TemporalAttNet(in_dim=out_channels, seq_len=hparams['motion_smo_win_size']) def forward(self, x): """ x: [B, 3, H=512, W=512] or [B, 3, T, H, W] return: plane: [B, 96, H=256, W=256] """ # if hparams['use_motion_smo_net']: # assert x.ndim == 5 # x = rearrange(x, "n c t h w -> (n t) c h w", t=self.motion_smo_win_size) x = self.prenet(x) feats = self.mix_vit(x) fused_feat = self.fuse_head(feats) planes = self.to_plane_cnn(fused_feat) # if hparams['use_motion_smo_net']: # planes = rearrange(planes, "(n t) c h w -> n c t h w", t=self.motion_smo_win_size) # planes = self.smo_net(planes) planes = planes.view(len(planes), 3, -1, planes.shape[-2], planes.shape[-1]) planes_xy = planes[:,0] planes_xy = torch.flip(planes_xy, [2]) planes_xz = planes[:,1] planes_xz = torch.flip(planes_xz, [2]) planes_zy = planes[:,2] planes_zy = torch.flip(planes_zy, [2, 3]) planes = torch.stack([planes_xy, planes_xz, planes_zy], dim=1) # [N, 3, C, H, W] return planes # from modules.hidenerf.new_modules.texture2plane_parser import Texture2PlaneParser # class SegFormerTexture2PlaneBackbone(nn.Module): # def __init__(self, mode='b1'): # super().__init__() # mode2cls = { # 'b0': mit_b0, # 'b1': mit_b1, # 'b2': mit_b2, # 'b3': mit_b3, # 'b4': mit_b4, # 'b5': mit_b5, # } # self.mode = mode # self.prenet = Conv2dLayer(5, 3, 1) # self.tex2plane_parser = Texture2PlaneParser() # self.mix_vit = mode2cls[mode]() # self.fuse_head = SegFormerHead(mode) # if hparams.get("new_tex_mode", False) is True: # self.to_plane_cnn1 = nn.Sequential(*[ # nn.Conv2d(in_channels=256, out_channels=256, kernel_size=3, stride=1, padding=1), # nn.LeakyReLU(negative_slope=0.01, inplace=True), # nn.UpsamplingBilinear2d(scale_factor=2.), # ]) # self.to_plane_cnn2 = nn.Sequential(*[ # nn.Conv2d(in_channels=256*3, out_channels=256, kernel_size=3, stride=1, padding=1), # nn.LeakyReLU(negative_slope=0.01, inplace=True), # nn.Conv2d(in_channels=256, out_channels=256, kernel_size=3, stride=1, padding=1), # nn.LeakyReLU(negative_slope=0.01, inplace=True), # nn.Conv2d(in_channels=256, out_channels=96, kernel_size=3, stride=1, padding=1) # ]) # else: # self.to_plane_cnn = nn.Sequential(*[ # nn.Conv2d(in_channels=256, out_channels=256, kernel_size=3, stride=1, padding=1), # nn.LeakyReLU(negative_slope=0.01, inplace=True), # nn.Conv2d(in_channels=256, out_channels=256, kernel_size=3, stride=1, padding=1), # nn.LeakyReLU(negative_slope=0.01, inplace=True), # nn.Conv2d(in_channels=256, out_channels=256, kernel_size=3, stride=1, padding=1), # nn.LeakyReLU(negative_slope=0.01, inplace=True), # nn.UpsamplingBilinear2d(scale_factor=2.), # nn.Conv2d(in_channels=256, out_channels=32, kernel_size=3, stride=1, padding=1), # ]) # def forward(self, x, idx_pixel_to_plane): # """ # x: [B, 3, H=512, W=512] # return: # plane: [B, 96, H=256, W=256] # """ # feats = self.mix_vit(x) # fused_feat = self.fuse_head(feats) # [B, 256, 128, 128] # if hparams.get("new_tex_mode", False) is True: # fused_feat = self.to_plane_cnn1(fused_feat) # [B, 96, 256, 256] # fused_feat = fused_feat.unsqueeze(1).repeat([1, 3, 1, 1, 1]) # [B, 3, 96, 256, 256] # tex_plane = self.tex2plane_parser(fused_feat, idx_pixel_to_plane) # [B, 3, 96, 256, 256] # tex_plane = rearrange(tex_plane, "n k c h w -> n (k c) h w") # [B, 3*96, 256, 256] # tex_plane = self.to_plane_cnn2(tex_plane) # [B, 96, 256, 256] # tex_plane = rearrange(tex_plane, "n (k c) h w -> n k c h w", k=3, c=32) # [B, 3*96, 256, 256] # else: # fused_feat = self.to_plane_cnn(fused_feat) # [B, 32, 256, 256] # fused_feat = fused_feat.unsqueeze(1).repeat([1, 3, 1, 1, 1]) # [B, 3, 32, 256, 256] # tex_plane = self.tex2plane_parser(fused_feat, idx_pixel_to_plane) # [B, 3, 32, 256, 256] # return tex_plane if __name__ == '__main__': import tqdm img2plane = SegFormerTexture2PlaneBackbone() img2plane.cuda() x = torch.randn([4, 3, 512, 512]).cuda() idx = torch.randint(low=0, high=128*128, size=[4, 3, 256*256]).cuda() for _ in tqdm.trange(100): y = img2plane(x, idx) print(" ")