import torch
import torch.nn as nn
import timm
from transformers import PreTrainedModel, PretrainedConfig

class HarpoonConfig(PretrainedConfig):
    model_type = "harpoon"
    
    def __init__(self, num_classes=1, resolution=544, **kwargs):
        super().__init__(**kwargs)
        self.num_classes = num_classes
        self.resolution = resolution

class HarpoonModel(PreTrainedModel):
    config_class = HarpoonConfig
    
    def __init__(self, config):
        super().__init__(config)
        self.num_classes = config.num_classes
        self.resolution = config.resolution
        
        # ConvNeXt Small backbone
        self.backbone = timm.create_model(
            'convnext_small',
            pretrained=False,
            num_classes=0,
            global_pool='',
            in_chans=3
        )
        
        # Harpoon Core detection head
        self.harpoon_core = nn.Sequential(
            nn.Conv2d(768, 256, 3, padding=1),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True),
            nn.Conv2d(256, 128, 3, padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True),
            nn.Conv2d(128, (5 + self.num_classes) * 3, 1)
        )
        
        self._initialize_harpoon_core()
    
    def _initialize_harpoon_core(self):
        """Initialize Harpoon Core detection head"""
        for m in self.harpoon_core.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
                if m.bias is not None:
                    nn.init.constant_(m.bias, 0)
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)
    
    def forward(self, x):
        # ConvNeXt backbone feature extraction
        features = self.backbone(x)
        
        # Harpoon Core detection
        detections = self.harpoon_core(features)
        
        return detections