Pppd515mp4 Extra Quality -

# 6️⃣ Final embedding embed = self.embed_head(pooled) # (B, out_dim) embed = nn.functional.normalize(embed, p=2, dim=-1) # L2‑norm

# ---------------------------------------------------------------------- # 2️⃣ BACKBONE DEFINITIONS # ---------------------------------------------------------------------- class FrameCNN(nn.Module): """ 2‑D CNN that produces a 1024‑D per‑frame descriptor. Using EfficientNet‑B4 (pre‑trained on ImageNet21k → strong texture sensitivity). """ def __init__(self): super().__init__() self.backbone = torchvision.models.efficientnet_b4(pretrained=True).features self.pool = nn.AdaptiveAvgPool2d(1) # -> (B, C, 1, 1) self.out_dim = 1792 # EfficientNet‑B4 final channel count pppd515mp4 extra quality