Amazon Product Descriptions With Generated Text For Vision-Language Models

A modified version of the Amazon Multimodal Product dataset, slimmed for training multimodal LLMs. The dataset includes product descriptions generated using the Gemini Flash model. It was created by philschmid and last updated in September 2024.

Use Cases

• Training a model to generate product descriptions from product images using the 'description' field.
• Fine-tuning a VLM to align visual product features with textual attributes in the dataset.
• Benchmarking multimodal retrieval systems on e-commerce product data with generated text.
• Developing product recommendation systems that leverage both visual and generated descriptive features.

Strengths

• Dataset is specifically slimmed and modified for efficient training of multimodal LLMs.
• Textual 'description' field was generated using a state-of-the-art model (Gemini Flash).

Limitations

• Unknown row count, column structure, and sample size prevents assessment of scale.
• Reliance on generated text for descriptions may introduce model-specific biases or errors.
• Lack of original metadata limits understanding of data modifications and potential information loss.

Provenance

Source

Modified from the 'bprateek/amazon_product_description' dataset on Hugging Face.

Collection Method

Descriptions were generated using the Gemini Flash model.

Time Range

null

Freshness

Last updated in September 2024.

Geography

null