ImageNet 1K Dataset at 128x128 Resolution

benjamin-paine repackaged the ImageNet 1K dataset into a 128x128 pixel format. Images were center-cropped and resampled to this uniform size using Lanczos interpolation. The repack was last updated on September 15, 2024.

Use Cases

• Train image classification models using the 128x128 pixel images and their associated 1,000-class labels.
• Benchmark model performance on a standardized, lower-resolution version of the ImageNet validation set.
• Use the pre-processed 128x128 images for knowledge distillation or student model training.
• Fine-tune pre-trained models on a uniformly sized dataset to reduce computational overhead.

Strengths

• Derived from the widely recognized ImageNet 1K benchmark dataset.
• All images are uniformly pre-processed to 128x128 resolution, ensuring consistent input dimensions.
• Images were rescaled using Lanczos resampling, a high-quality interpolation method.

Limitations

• Original image details and aspect ratios are lost due to center-cropping and downscaling.
• Dataset size, row count, and class distribution specifics are not provided in the input.
• The dataset is a derivative repack; license and original provenance details are not specified here.

Provenance

Source

ILSVRC/ImageNet-1k.

Collection Method

Repackaged from the original dataset with center-cropping and Lanczos resampling to 128x128 pixels.

Time Range

null

Freshness

Repack last updated in September 2024.

Geography

null