Loading custom object detection data

vision
1

I’m wanting to train a SSD-Mobilenet model using my own dataset.

My dataset is labelled, below is the structure of my data;

Dataset

JPEGImages

0001.jpeg
0002.jpeg…

Annotations

0001.XML
0002.XML

Almost all tutorials i can find either use built in datasets or datasets containing a csv file.

Any ideas on how i can load the above structure into pytorch,I’ll be using torchvision.

Thanks

2

You can get a list of files using os.listdir. Then, you can write a custom dataloader like this:

class data_gen(torch.utils.data.Dataset):
    def __init__(self, files):
        
        self.files = files
        
    def __getitem__(self, I):

        #read the images here, convert to torch format and return


    def __len__(self): 
        
        return <number of images>

Then, use an instance like train_dl = data_gen(image_files) and finally use

train_loader = torch.utils.data.DataLoader(
        train_dl, batch_size= , shuffle=True, num_workers= , pin_memory=True)
3

Thanks for a prompt reply. I’ll give it a shot now.

4

Below is teh example code i’m trying to adapt, but the issue is i don’t have a csv file my annoattaions are simple in teh annotation folder as .xml???

class data_Gen(torch.utils.data.Dataset):
def init(self,csv_file,root_dir,transform):
self.annotations=pd.read_csv(csv_file)
self.root_dir=root_dir
self.transform=transform

def __getitem__(self,I):
    img_path=os.path.join(self.root_dir,self.annotations.iloc[index,0])
    image=io.imread((img_path))
    y_label=torch.tensor((int(self.annotations.iloc[index,1])))
    
    if self.transform:
        image=self.transform(image)
        
    return(image,y_label)
5

You can read xml files in python. Look for library that does it and get the annotations from there and return it as y_label

6

Thanks and really sorry for the late reply, I just noticed your answer