TheoBourdais / ImageNetSubmission

This is a simple example of how to prepare an ImageNet submission to the evaluation server.

Simple code for preparing a submission to the ImageNet evaluation server for the test set

This code is a simple example of how to prepare an ImageNet submission to the evaluation server. A notebook is provided that shows how to use the code to prepare a submission. The folder src contains the code for preparing the submission, as well as the file idx_to_ILSVRC_ID.csv which maps the ImageNet class index (that are given automatically when using torchvision.datasets.ImageFolder) to the ILSVRC ID needed for the submission. It assumes you have already downloaded the ImageNet test dataset, which can be found here. Once the submission is prepared, you can submit it to the ImageNet evaluation server to get the results.

Note: There are more details on how to obtain the idx_to_ILSVRC_ID.csv file in the README of the src folder.

Usage

You may open the notebook example.ipynb and modify it to prepare a submission. The notebook is self-contained and should be easy to follow. You can also directly use the code below

import torch
from torch.utils.data import DataLoader
import timm
import os
from src.utils import TestDataset, get_test_submission


device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# Prepare the data loaders and model

model = timm.create_model("vit_base_patch16_224", pretrained=True)

data_config = timm.data.resolve_model_data_config(model)
preprocess = timm.data.create_transform(**data_config, is_training=False)
model.to(device)


test_dataset = TestDataset("./imagenet/test/", transform=preprocess)
loader = DataLoader(test_dataset, batch_size=100, shuffle=False)

# Prepare the submission

submission = get_test_submission(model, loader, device)
submission.to_csv("submission.txt", index=False, header=False, sep=" ")

License

Public domain.