models/heads.py

import torch
import torch.nn as nn
import torch.nn.functional as F


class CropFCNHead(nn.Module):
    """
    AgriFM segmentation head - faithful to GitHub implementation.
    Conv(embed -> embed//2) + ReLU + Conv(embed//2 -> num_classes)
    """

    def __init__(self, embed_dim, num_classes):
        super().__init__()
        self.embed_dim = embed_dim
        self.num_classes = num_classes
        self.head = nn.Sequential(
            nn.Conv2d(embed_dim, embed_dim // 2, kernel_size=3, stride=1, padding=1),
            nn.ReLU(inplace=True),
            nn.Conv2d(embed_dim // 2, num_classes, kernel_size=1, stride=1, padding=0),
        )

    def forward(self, x):
        """
        x: (B, embed_dim, H, W)
        returns: (B, num_classes, H, W)
        """
        return self.head(x)