Dataset Viewer
Auto-converted to Parquet Duplicate
image
imagewidth (px)
384
384
text
stringlengths
24
90
a woman wearing a yellow sequin dress
a woman in a leopard print dress
a woman wearing a green dress with a strap
a pink dress with a white teddy bear
a woman wearing a black and white plaid dress
a woman wearing a yellow floral print dress
a woman in a brown dress posing on a white couch
a woman wearing a green floral print dress
a woman in a dress posing for the camera
a pink dress with the words perfectly imperfect on it
a woman wearing a pink dress
a woman wearing a green dress and black gloves
a woman wearing a green pleaed dress
a woman sitting on a blanket with a cup of tea
a woman in a blue dress leaning against a wooden wall
a woman wearing a blue shirt and shorts
a woman wearing a blue dress with a large bow
pink spaghetti strappy dress
a woman in a pink and yellow floral print dress
a woman wearing a white lace dress
a woman in a dress sitting on a chair
a woman in a purple dress
a woman wearing a black sequin dress
a woman in a red velvet gown
a woman wearing a brown dress with a long sleeve and a high neckline
a woman wearing a maroon dress
a woman wearing a black dress with a cleavager
a woman in a pink dress sitting on a chair
a woman wearing a striped dress
a woman walking on the beach wearing a brown and black dress
a woman wearing a straw hat and a dress
a woman wearing a pink dress with a long sleeve
a woman in a yellow turtle neck dress
a woman in a blue top and orange skirt
a woman in a black and tan dress on the beach
a blue and white dress with a white collar
a woman in a blue dress and hat
a white dress with black and white spots
a woman in a long dress and boots
a woman wearing a white dress with wavy waves
a woman in a purple sei gown
a woman in a white dress with a diamond belt
a woman in a black gown with a flower bouquet
a woman wearing a red dress and black shoes
a woman wearing a pink dress and sunglasses
a woman wearing a floral print dress
a woman in a black dress sitting on a couch
a woman in a green dress leaning against a wall
a woman wearing a burgundy dress with a white purse
a woman sitting in a chair wearing a black dress
a woman wearing a floral print dress
a woman wearing a red velvet dress
a woman wearing a blue dress with white trims
a woman in a white and purple dress
a woman in a black dress standing on a sidewalk
a woman in a white dress posing for the camera
a woman wearing a brown dress with long sleeves and a gold chain
a woman in a brown dress holding a cup
a woman wearing a brown dress with a long sleeve
a woman in a dress and sunglasses is taking a self self self self self self self self self
a woman in a brown dress standing in front of a door
a woman wearing a green and white floral print dress
a woman in a floral print dress standing on the beach
a white dress with a tie around the neck and straps
a woman wearing a navy and white dress
a woman in a yellow shirt dress is posing
a woman in a purple sequin dress
a woman wearing a green dress with a brown bag
a woman wearing a striped dress with a straw hat
a woman wearing a black dress with a long sleeve and a high slit
a woman wearing a pink dress with a white clutch
a maroon t - shirt with the words perfectly perfect on it
a woman wearing a white top and red floral skirt
a woman wearing a leopard print dress
a woman wearing a burgundy dress
a woman in a green dress and sandals
a woman sitting on the floor wearing a black top and floral skirt
a woman wearing a green sweater dress
a woman wearing a hat and coat
a woman wearing a grey striped dress with a black belt
a woman in a dress and straw hat on the beach
a purple dress with a sei sei sei sei sei sei sei
a woman in a purple dress with a large cleavage
a woman wearing a black dress with lace detailing
a woman wearing a brown polka print dress
a woman wearing a green dress
a woman in a black dress posing for the camera
a woman in an orange floral print swimsuit
a woman in a blue dress standing in front of a mirror
a woman in a green dress
a woman wearing a black dress with a high neck and long sleeves
a woman in a pink dress sitting on a stone wall
a woman wearing a green dress with a cutout back
a woman wearing a blue dress with a white purse
a woman wearing a green floral print shirt dress
a woman wearing a black and white dress
a woman in an orange dress posing for the camera
a woman in a brown dress holding a white purse
a woman wearing a green velvet dress
a woman in a pink dress is holding a purse

Fashion Image Captions (BLIP-generated)

A derived dataset based on adirik/fashion_image_caption-100, with image captions regenerated using the BLIP image captioning model.

What was done

The original dataset contains fashion product images paired with human-written descriptions. Using the Hugging Face datasets .map() function, the original captions were replaced with captions automatically generated by BLIP (Salesforce/blip-image-captioning-base).

Dataset details

Attribute Value
Source dataset adirik/fashion_image_caption-100
Captioning model Salesforce/blip-image-captioning-base
Domain Fashion / clothing
Size 100 image-caption pairs
Task Image captioning

Processing pipeline

Captions were regenerated by applying a custom replace_caption() function via the datasets .map() method:

from transformers import BlipProcessor, BlipForConditionalGeneration

processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")

def replace_caption(data):
    # Preprocess image
    inputs = processor(images=data["image"], return_tensors="pt")
    # Run inference with torch.no_grad() and the generate() method
    with torch.no_grad():
        output = model.generate(**inputs)
    # Decode and update the text feature
    data["text"] = processor.decode(output[0], skip_special_tokens=True)
    return data

new_dataset = dataset.map(replace_caption)

This pattern — wrapping preprocessing and model inference inside a single function and applying it across the full dataset via .map() — is a standard HF datasets workflow for building multimodal data pipelines.

Note: Processing all 100 samples on CPU (2 cores, 4GB RAM) took approximately 30 minutes. GPU acceleration is recommended for larger datasets.

Potential uses

  • Comparing automatically generated captions (BLIP) vs. original human-written descriptions for fashion images
  • Lightweight testbed for image captioning experiments in the fashion domain

Notes

This dataset was created as a portfolio exercise while working through the Hugging Face NLP/multimodal ecosystem — specifically practising multimodal pipelines, the datasets .map() function, and Hub dataset publishing workflows.

Downloads last month
5