그동안의 개념이 혼합되어 쓰였었는데, 이 논문을 통해 Natural language supervision으로 명칭하게 되었다.
but, 이러한 장점에도 ConVIRT와 VirTex는 10만 개에서 20만 개정도의 이미지를 학습했다.
*downstream task: 최종적으로 해결하고자 하는 작업, 여기서는 내가 최종적으로 만들고자 하는 모델을 의미한다. upstream task로는 pre-trained model이라고 생각하면 될 듯하다.
Pre-train method로 contrasive learning을 이용했고, text-image pair 이용해 pre-train 시켰다.
VirTex 사용
Contrastive Learning
# image_encoder - ResNet or Vision Transformer
# ViT-L/14가 가장 좋은 성능을 보여 모든 실험의 backbone으로 사용됨
# text_encoder - CBOW or Text Transformer
# Text Transformer 가 사용된듯.. Sparse Transformer 사용
# I[n, h, w, c] - minibatch of aligned images
# T[n, l] - minibatch of aligned texts
# W_i[d_i, d_e] - learned proj of image to embed
# W_t[d_t, d_e] - learned proj of text to embed
# t - learned temperature parameter
# learnable parmeter for scaling
# extract feature representations of each modality
I_f = image_encoder(I) # [n, d_i]
T_f = text_encoder(T) # [n, d_t]
# joint multimodal embedding [n, d_e]
I_e = l2_normalize(np.dot(I_f, W_i), axis=1)
T_e = l2_normalize(np.dot(T_f, W_t), axis=1)
# linear layer를 이용해서 임베딩 스페이스로 각각을 보내고, dot product!
# 이후 둘을 내적해서 유사도 확인할 것이니, l2 normalization으로 크기 맞춰주기
# scaled pariwise cosine similarities [n, n]
logits = np.dot(I_e, T_e.T) * np.exp(t)
# I_e와 전치된 T_e를 내적하고, e의 t(스케일링 파라미터)승을 곱해 코사인 유사도 확인
# symmetric loss function
labels = np.arange(n)
loss_i = cross_entropy_loss(logits, labels, axis=0)
loss_t = cross_entropy_loss(logits, labels, axis=1)
# cross entropy loss로 최적화시켜줌
loss = (loss_i + loss_t) / 2 # 두 모달리티 loss 값의 평균을 loss로 봄
문장 형태로 prompting을 진행한다.A photo of a {object}.
ViT-L/14
가 가장 좋은 성능을 보여 모든 실험의 backbone으로 쓰임CLIP 모델의 레포지토리, model.py 코드 보면서 코드 정리 진행
class CLIP(nn.Module):
def __init__(self,
embed_dim: int,
# vision
image_resolution: int,
vision_layers: Union[Tuple[int, int, int, int], int],
vision_width: int,
vision_patch_size: int,
# text
context_length: int,
vocab_size: int,
transformer_width: int,
transformer_heads: int,
transformer_layers: int
):
super().__init__()
self.context_length = context_length
if isinstance(vision_layers, (tuple, list)):
vision_heads = vision_width * 32 // 64
self.visual = ModifiedResNet(
layers=vision_layers,
output_dim=embed_dim,
heads=vision_heads,
input_resolution=image_resolution,
width=vision_width
)
else:
vision_heads = vision_width // 64
self.visual = VisionTransformer(
input_resolution=image_resolution,
patch_size=vision_patch_size,
width=vision_width,
layers=vision_layers,
heads=vision_heads,
output_dim=embed_dim
)
@ property
def encode_image(self, image):
return self.visual(image.type(self.dtype))
class ModifiedResNet(nn.Module):
"""
A ResNet class that is similar to torchvision's but contains the following changes:
- There are now 3 "stem" convolutions as opposed to 1, with an average pool instead of a max pool.
- Performs anti-aliasing strided convolutions, where an avgpool is prepended to convolutions with stride > 1
- The final pooling layer is a QKV attention instead of an average pool
"""
def __init__(self, layers, output_dim, heads, input_resolution=224, width=64):
super().__init__()
self.output_dim = output_dim
self.input_resolution = input_resolution
# the 3-layer stem
self.conv1 = nn.Conv2d(3, width // 2, kernel_size=3, stride=2, padding=1, bias=False)
self.bn1 = nn.BatchNorm2d(width // 2)
self.relu1 = nn.ReLU(inplace=True)
self.conv2 = nn.Conv2d(width // 2, width // 2, kernel_size=3, padding=1, bias=False)
self.bn2 = nn.BatchNorm2d(width // 2)
self.relu2 = nn.ReLU(inplace=True)
self.conv3 = nn.Conv2d(width // 2, width, kernel_size=3, padding=1, bias=False)
self.bn3 = nn.BatchNorm2d(width)
self.relu3 = nn.ReLU(inplace=True)
self.avgpool = nn.AvgPool2d(2)
# residual layers
self._inplanes = width # this is a *mutable* variable used during construction
self.layer1 = self._make_layer(width, layers[0])
self.layer2 = self._make_layer(width * 2, layers[1], stride=2)
self.layer3 = self._make_layer(width * 4, layers[2], stride=2)
self.layer4 = self._make_layer(width * 8, layers[3], stride=2)
embed_dim = width * 32 # the ResNet feature dimension
self.attnpool = AttentionPool2d(input_resolution // 32, embed_dim, heads, output_dim)
def _make_layer(self, planes, blocks, stride=1):
layers = [Bottleneck(self._inplanes, planes, stride)]
self._inplanes = planes * Bottleneck.expansion
for _ in range(1, blocks):
layers.append(Bottleneck(self._inplanes, planes))
return nn.Sequential(*layers)
def forward(self, x):
def stem(x):
x = self.relu1(self.bn1(self.conv1(x)))
x = self.relu2(self.bn2(self.conv2(x)))
x = self.relu3(self.bn3(self.conv3(x)))
x = self.avgpool(x)
return x
x = x.type(self.conv1.weight.dtype)
x = stem(x)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
x = self.attnpool(x)
return x
class VisionTransformer(nn.Module):
def __init__(self, input_resolution: int, patch_size: int, width: int, layers: int, heads: int, output_dim: int):
super().__init__()
self.input_resolution = input_resolution
self.output_dim = output_dim
self.conv1 = nn.Conv2d(in_channels=3, out_channels=width, kernel_size=patch_size, stride=patch_size, bias=False)
scale = width ** -0.5
self.class_embedding = nn.Parameter(scale * torch.randn(width))
self.positional_embedding = nn.Parameter(scale * torch.randn((input_resolution // patch_size) ** 2 + 1, width))
self.ln_pre = LayerNorm(width)
self.transformer = Transformer(width, layers, heads)
self.ln_post = LayerNorm(width)
self.proj = nn.Parameter(scale * torch.randn(width, output_dim))
def forward(self, x: torch.Tensor):
x = self.conv1(x) # shape = [*, width, grid, grid]
x = x.reshape(x.shape[0], x.shape[1], -1) # shape = [*, width, grid ** 2]
x = x.permute(0, 2, 1) # shape = [*, grid ** 2, width]
x = torch.cat([self.class_embedding.to(x.dtype) + torch.zeros(x.shape[0], 1, x.shape[-1], dtype=x.dtype, device=x.device), x], dim=1) # shape = [*, grid ** 2 + 1, width]
x = x + self.positional_embedding.to(x.dtype)
x = self.ln_pre(x)
x = x.permute(1, 0, 2) # NLD -> LND
x = self.transformer(x)
x = x.permute(1, 0, 2) # LND -> NLD
x = self.ln_post(x[:, 0, :])
if self.proj is not None:
x = x @ self.proj
return