Missing MNIST Dataset Files for TensorRT Sample Code

Issue Overview

Users are encountering an error when attempting to execute the sample.py file from the /usr/src/tensorrt/samples/python/int8_caffe_mnist/ directory on the Nvidia Jetson Orin Nano Dev board. The error indicates that required MNIST dataset files are missing from the expected locations. Specifically, the following files are not present:

• "t10k-images-idx3-ubyte"
• "t10k-labels-idx1-ubyte"
• "train-images-idx3-ubyte"

These files are essential for running the TensorRT sample code that demonstrates Int8 mode execution of a Caffe MNIST network. The absence of these files prevents the sample from running successfully, impacting users’ ability to test and learn from the provided example.

Possible Causes

1. Incomplete Dataset Download: The MNIST dataset files may not have been downloaded or placed in the correct directory during the initial setup.

2. Incorrect File Permissions: The required files might exist but have incorrect permissions, preventing the script from accessing them.

3. Corrupted or Incomplete Installation: The TensorRT samples installation could be incomplete or corrupted, resulting in missing dataset files.

4. Outdated Documentation: The README file or documentation might not provide clear instructions on obtaining the necessary dataset files.

5. File Path Discrepancy: The script might be looking for the files in an incorrect location due to a mismatch between the expected and actual file paths.

Troubleshooting Steps, Solutions & Fixes

1. Verify File Locations:

   • Check the /usr/src/tensorrt/data/mnist/ directory to confirm if the required files are present.
   • Ensure you have the necessary permissions to access these directories.

2. Download Missing Files:

   • Access the MNIST dataset from the official website: http://yann.lecun.com/exdb/mnist/
   • Download the following files:
     • http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz
     • http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz
     • http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz

3. Extract and Place Files:

   • Extract the downloaded .gz files.
   • Move the extracted files to the /usr/src/tensorrt/data/mnist/ directory.

4. Set Correct Permissions:

   • Ensure the files have the correct read permissions:

     sudo chmod 644 /usr/src/tensorrt/data/mnist/*
     

5. Update TensorRT Samples:

   • If you suspect an incomplete installation, consider reinstalling or updating the TensorRT samples:

     sudo apt-get update
     sudo apt-get install --reinstall tensorrt-samples
     

6. Verify Script Paths:

   • Open the sample.py file and check if the file paths are correctly specified.
   • Ensure the script is looking in the right directory for the dataset files.

7. Run the Sample:

   • After placing the files in the correct location, run the sample again:

     python3.6 sample.py -d /usr/src/tensorrt/data
     

8. Check for Updates:

   • Visit the NVIDIA Developer website to check for any updates or known issues related to the TensorRT samples on the Jetson Orin Nano.

9. Consult Documentation:

   • Review the README file in the sample directory for any additional setup steps or requirements.

By following these steps, users should be able to resolve the issue of missing MNIST dataset files and successfully run the TensorRT sample code on their Nvidia Jetson Orin Nano Dev board.