RF100-VL Dataset Interface

1.1.0 · active · verified Fri Apr 17

`rf100vl` is a Python library that provides a convenient interface for the RF100-VL dataset, specifically designed for research in multi-modal learning and understanding. It handles the downloading, caching, and access of the dataset's image-caption pairs, allowing users to easily integrate it into their machine learning pipelines. The current stable version is 1.1.0, and the project appears to be in maintenance with occasional minor updates.

Common errors

```
ModuleNotFoundError: No module named 'rf100vl.RF100VL'
```
cause The import statement incorrectly assumes the `RF100VL` class is directly under the `rf100vl` package, rather than its nested module.

fix Adjust your import statement to `from rf100vl.rf100vl import RF100VL`.
```
requests.exceptions.ConnectionError: ('Connection aborted.', ConnectionResetError(104, 'Connection reset by peer'))
```
cause An issue with network connectivity, firewall restrictions, or proxy configuration is preventing the download of dataset files from the server.

fix Check your internet connection, verify any firewall or proxy settings, and try the operation again. Ensure your system can reach `s3.eu-central-1.amazonaws.com`.
```
FileNotFoundError: [Errno 2] No such file or directory: '/some/invalid/path/annotations.json'
```
cause The `root_dir` provided during `RF100VL` initialization does not exist, or the dataset files were not successfully downloaded/extracted into that location.

fix Ensure the directory specified by `root_dir` exists by calling `os.makedirs(root_dir, exist_ok=True)` before creating the `RF100VL` instance. Also, confirm that `download=True` is set and that the download completed without errors.

Warnings

gotcha The RF100-VL dataset is substantial in size (multiple gigabytes). Ensure your system has sufficient free disk space and a stable, high-bandwidth internet connection before attempting the initial download. The download process can be lengthy.
Fix: Verify available disk space and network stability. The `tqdm` progress bar will indicate download status, but be prepared for a long wait.
gotcha The primary class `RF100VL` is located within the `rf100vl.rf100vl` module, not directly under the `rf100vl` package. A common mistake is to omit the inner `rf100vl` in the import path, leading to an `ImportError`.
Fix: Always use `from rf100vl.rf100vl import RF100VL` for correct importation.
gotcha The `root_dir` parameter for `RF100VL` specifies where the dataset files are stored. If this directory does not exist, the library might raise a `FileNotFoundError` or attempt to create it without proper permissions, leading to issues. Subsequent file access will also fail if the path is invalid.
Fix: Always ensure the `root_dir` exists by calling `os.makedirs(root_dir, exist_ok=True)` before initializing the `RF100VL` class.

Install

pip install rf100vl Install stable version

Imports

RF100VL
wrong:
```
from rf100vl import RF100VL
```
correct:
```
from rf100vl.rf100vl import RF100VL
```
The main dataset class is nested one level deeper within the `rf100vl` package structure, requiring `rf100vl.rf100vl` as the module path.

Quickstart

This quickstart demonstrates how to initialize the `RF100VL` dataset, automatically downloading it to a specified `root_dir` if it's not already present. It then shows how to access an individual item, which provides a PIL Image and its corresponding text caption.

import os
from rf100vl.rf100vl import RF100VL

# Define a directory for the dataset; it will be created if it doesn't exist.
# Using an environment variable or a default path for flexibility.
data_root = os.environ.get('RF100VL_DATA_ROOT', './rf100vl_data')
os.makedirs(data_root, exist_ok=True)

try:
    # Initialize the dataset. Set download=True to fetch if not present.
    # This can take significant time and disk space.
    dataset = RF100VL(root_dir=data_root, split='train', download=True)

    print(f"\nSuccessfully loaded RF100VL dataset with {len(dataset)} items in '{data_root}'.")

    # Access a sample item (e.g., the first one)
    sample_item = dataset[0]
    image = sample_item['image'] # A PIL Image object
    caption = sample_item['caption'] # A string caption

    print(f"\nFirst item details:")
    print(f"  Caption: '{caption[:100]}...' ")
    print(f"  Image type: {type(image)}, size: {image.size}, mode: {image.mode}")

    # Further processing (e.g., transforming image, tokenizing caption) would go here.

except Exception as e:
    print(f"\nAn error occurred during dataset initialization or access: {e}")
    print("Please ensure you have network access, sufficient disk space, and correct permissions for the data_root directory.")

view raw JSON →