dog-localization / models.py
sadjava's picture
updated model
a741f39
raw
history blame
1.06 kB
import torch.nn as nn
from torchvision.models import resnet18, ResNet18_Weights
import torch.nn.functional as F
import torch
class Model(nn.Module):
def __init__(self):
super().__init__()
self.feature_extractor = resnet18(weights=ResNet18_Weights)
in_channels = self.feature_extractor.fc.in_features
self.feature_extractor.fc = nn.Identity()
# Output is a vector of dimension 1 + 4
# 1 for probability of belonging to any class
# 4 for bounding box of object that is presented (if no object is presented i. e. the probability < a threshold, any 4 numbers)
self.fc_prob = nn.Sequential(
nn.Linear(in_channels, 512),
nn.Linear(512, 1)
)
self.fc_bbox = nn.Sequential(
nn.Linear(in_channels, 512),
nn.Linear(512, 4)
)
def forward(self, x):
pred_prob = torch.sigmoid(self.fc_prob(self.feature_extractor(x)))
pred_bbox = self.fc_bbox(self.feature_extractor(x))
return (pred_prob, pred_bbox)