# sam2_model_stub.py

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

class SAM2Hierarchical(nn.Module):
    def __init__(self, num_classes=6, in_channels=3, backbone="vit_b", freeze_backbone=True, use_cls_head=True):
        super().__init__()
        self.use_cls_head = use_cls_head

        # Minimal vision backbone stub (fake transformer or CNN)
        self.backbone = nn.Sequential(
            nn.Conv2d(in_channels, 64, kernel_size=3, stride=2, padding=1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True),
            nn.Conv2d(64, 128, kernel_size=3, stride=2, padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True)
        )

        # Segmentation head stub
        self.segmentation_head = nn.Sequential(
            nn.Conv2d(128, 64, kernel_size=3, padding=1),
            nn.ReLU(inplace=True),
            nn.Conv2d(64, num_classes, kernel_size=1)
        )

        # Optional classification head
        if self.use_cls_head:
            self.cls_head = nn.Linear(128, num_classes)

        if freeze_backbone:
            for param in self.backbone.parameters():
                param.requires_grad = False

    def forward(self, x):
        features = self.backbone(x)
        logits = self.segmentation_head(features)

        if self.use_cls_head:
            # Just return segmentation output; inference only cares about logits
            return logits

        return logits