Datasets:
uid
stringlengths 32
32
| blip2-cap
stringlengths 1
508
|
---|---|
e057914bf2263c053f0c1c91e2612eff | the logo of ganiti on a white background |
398b41a311a43707cb4b40b0143eeea0 | a 3d render of a metal ball |
f265c95351eb742f57fe65e1e65e0161 | 【電子書籍書き換え】【同配送】【電子書籍 |
c0bb73ff337c9569d982c21be1096679 | a husky dog playing with its toy |
964ee0c20c0cf32d2de63d2b59644b75 | e3 2013 e3 2013: the games industry's biggest announcements |
1b8abe265dd5c87a424246571ff04765 | indian express tamil - isanadu |
816fe75ab93c6e92099c7f5aaeb7fc43 | an image of a woman in a circle |
a72a26fd7c4f0c6880668e7795f7d584 | a large pile of sand in the middle of a desert |
550b206ab5630fdbe9ad8cace2cbe203 | industrial grade crane with crane safety equipment |
10bbdce1c3e58397d05fad6fb423f6f2 | an asian certificate on a black background |
289311e1d7dea77310e8ca5e47428846 | covid-19 manufacturing the fight continues |
30d80e6608c79bf6cea07303e8c00942 | vegetable bouillon in carton |
7bb69fc932a7af8d8298bcca8e96aba0 | an image of the man wearing a headset |
9c417beed4abcea9afa27e93bef6042b | many people are wearing the colors of the flags and holding them |
920c03849175c85ae80a7816fb722a60 | how to disable the advanced settings in the browser |
fd500c5b798d3b05c784c6bcd2dd3a90 | graffiti is on the side of the wall |
67d4d865c3bd4da002b02a9e60090e35 | 9780710556423 magic crochet by alexandra moore |
64234790d8b3d75366f60cafb588fcf0 | a painting of a palace with people and animals |
82c7f972c70a0e9a5e20f276163f71b6 | the green and white nike nike team jersey |
8ba2398bce39709324f9cf3be1f45134 | starter - stater - bmw - e30 / e36 |
126d27f398796a5b5078379db524ef05 | woman sitting in a cafe looking at laptop screen |
14b1145f23cc9287e32280a114b71aed | deer in the woods shower curtain |
01edbeaa5e032d2b6f5274e92a956a09 | a plastic class chair with an attached computer table |
6a588ba8a19d8cbba81e5b3fb2fb09e0 | a pair of black gloves with a black leather palm |
75796fdf2fdc11954f85b39665dfa30f | a spoon full of soup |
b25fc613e885830e6c6c450b1813db2e | money s - korea |
2caa9cab519cfc31aa2489fe2b64dd78 | camera ip hd/2m |
20f9ceba77cb2c05ac0b3f1401fb6347 | the broken floor around the door of an office is covered with food and debris |
e5faba8aa0d54053f1a6786818289537 | two people walk by a home with white fence and black wall |
0281d838eca94e63d983c4b1dc2c5209 | chumleei the children's books that are illustrated |
72a194ad6b7d34bc3ae9c4b1fc3e319a | the book cover for how to prune fruit trees |
66b05f28fd93ba24f6110a98669317a6 | yellow sunflower in the middle of a beautiful white sky |
402455025318adf0080e1b7833e64bb9 | a man with a white background and a man is in the portrait |
f68a1728792ae1c6098270618a17adb7 | a person leaning against a wall with their hand on their chin |
250baaa5fdff74a0640cf04334b0742d | the steps for making extra mac cakes |
3f765c8da2db528d08818bf5497a11a0 | vintage tikka jungle wallpaper, green |
eff8a474da1c4dd5a66def6dac1284cc | a green book with a gold color and yellow writing |
4531fc5a95a2e4fba4a7fe508fe8028d | men's reebo running shoes in blue |
5df883d661ac6faf613f8ac192bee08b | a couple rides in a vintage white vehicle down a road |
1fae9389b413f71ad3c1f89472ad9c4b | sky team kit for 2014 kit |
1a4f2305784b81105c0917debcccf775 | a very pretty shop with lots of items on display |
efc3a8dfc92210fa879a2d1ff77cca3a | the large stainless steel oven is sitting on a trailer |
82b260f4d7658d2dc16d0e991c4fa49f | a collection of colorful fabric is shown |
3086e0da0e0125a0e0e62874132a63ca | a jar of pink glitter on top of a white background |
3159168a3bb3e4b87c3d8115bfc7a207 | a car with the car's rear window open and the sunshade |
e483bc404d58941ab595308148657b7b | a console table with the shelves on top |
3f7eba3b55eb7710ff1273ec6231525b | sexta-feira, 27 de janeiro, ouviu um homem dizendo que o preto tinha se chumbado no mar |
6551d7480a36e787349541a4a3315a29 | manolis after sun lotion spray with blue cap 100ml |
2e7b52f0e228d222444ff7c3c79c2b92 | the internet of things in the constructional system |
4af63ba10c42e35049b72222020dcb73 | a white and blue image with the slogan everyday me |
3e73847be05d9d8e231c2cff6b16a19c | blue background with white dots and lines |
325b65d336b6cee0daa313a39009c4fa | a motorcycle parked on a dirt road next to a field |
bf5a0d8ca4cda179d24a786a2447e39b | the pursuit of happiness by anthony kane |
3c5db3d4f1ceef0ec723c19800d9d2ff | an red sign that reads photo coming soon |
f582154219b616b9d4e02e7b3251bbe5 | a blue and white umbrella standing upright with a handle |
8addecec005ab8a80f5a4b78c7ed9b69 | a man sitting in the driver seat of a car |
dcbb87b5e2c5ceb2b931502b581b2fc5 | the e logo is a flaming orange |
681cbac96f0f9ccfc99b69149f581cce | the person in blue is a man in a speech bubble on a white background |
5c6d2f9a509059cf6b3a862ad09d2773 | the best hydrating serums you can buy right now |
2864d7ffce1e109152945a4d03b17069 | telenor logo |
f659adee90eccd377e68c7d8164158ad | a poster from the movie 'the hunter' featuring the people on horseback |
513d1063c3014ba8336259fd0ba52a0d | a large duffel bag that is black and has straps in it |
1ffa1ce0b880a3d622c15a91b3b5d0a8 | a screen with a video about spacex and their mission |
a674fca9ac03bd53e6c1318316f89b0d | visit us - a good start, but need more than this |
986df01473f994a6d9d92ca3df22ed86 | 3d rendering of a reindeer jumping and jumping |
b6aad49a9d571f1e7acec86e2c9ec71d | fifa to allow gay, transsexual, other gay officials in 2017 |
2c0c387818d1ab267d1f3b1eb294944c | a view of ski villages from a snowy mountain |
ec6434da164f2087c20a9725b10f309b | a woman with long brown hair and wavy shoulder length hair |
0cff535c9eade70190dd4893ca32c564 | the new fiat 500x is driving on the highway at night |
40ace4c764e71a5f86befc3130698fbf | malaysia's new map reveals huge land expanse over seabed |
d51299345cac827a27f3c1ddc6f86d91 | a green pen with a white background |
eea8a9a8acf8ff6d52d75ad0102a8cd4 | a blue tin with many kitchen utensils in it |
411b175f4a6a0292c0038ee554ddb8e3 | etn2001-p rotary hammer |
75daeb9e4804e81d88189cddfc73ad22 | kitchenaid refrigerator door opener with 2 plastic handles |
b3aa4b62e27cc91012b8cdaba7390037 | a blouse with a bow on the front and buttons |
c9a253201971bb0f9b05bc596c178f8b | norse alphabet – the world's oldest writing system |
8474f8c6be6d268a4c44e0086aa13fc5 | a bathroom has purple walls and a white tub and toilet |
c94b7943d2e4d835c32d8b1a8fc831f6 | a safe with the electronic keypad and lock is in the safe |
a0f58d536875866adfedb336831ca259 | alessandro defilippi - il terremo delle ombra |
23a8bba64e1ab970355c07d47f31f557 | a pink ombre napkin with gold accents |
4d126b2b21eee91e2036e330fe3df74d | two metal dvd cases with two discs in each |
9d3e275ea4f5750370665f07d3dd5d5c | a screen shot of the windows 7 operating system |
1f4bee6e85226774391fb417c399e337 | the portable fan is connected to a black cord |
27a57069811c08e1ffe6498072bc097c | manhunter 13 comic books - covers, scans, photos in manhunter comic books - covers, scans, gallery |
33f93e51ef9954053be7accbb298feda | two people are flying with their arms out |
05847d78db73006379fba186e5802913 | a camera with a white background |
ee7a3794ad8a7c5d751734d36a81a016 | the pool and grassy lawn in front of the ocean in this beautiful villa |
64ea7b311a78a4265d55820ba8d95481 | the occupation xbox one game |
55598b081be3e7296c7c76dc14213ea0 | pink satin leopard skin ruffled mini dress r4 |
415619cea744ae40a8a1de697143ef45 | sunglasses sunglasses women sunglasses glasses polarized glasses blue lens sunglasses women sunglasses polarized glasses |
e887a6bb0ec40c5c5b6f5d41a028b8a9 | four men standing on a stage with their medals |
9cd82ba7c41cabb637b23381d50bf9fe | the dinosaur world by mylisa larsen |
c3dcbe680849971f862af3ab735b9665 | two men pose with a stuffed penguin in the middle of a room |
9c0d5400e5a388635963406c440cc620 | a yellow award ribbon with a smiling face |
6d5c2ee3856ee112ff537d103f555031 | an orange store with a sign that says orange in it |
501e86af7de55cf924f55f93553b0834 | the lion king lion king lion king lion king lion king lion king lion king lion king king lion king 1 |
702864b5b7a1e4ae72e85abc8278a95d | two potatoes with butter are on top of a tray |
35aa7540657edc443078d9797ec324a3 | an old engine part showing the number for the vehicle |
e33a1efe992c25f6cfcb1ca24d108b3a | a metal railing on the steps outside of a house |
4e3f45c612f9e6ad775145a78c7b1e13 | two chairs with plastic back |
Dataset Card for DataComp_large_pool_BLIP2_captions
Dataset Summary
Supported Tasks and Leaderboards
We have used this dataset for pre-training CLIP models and found that it rivals or outperforms models trained on raw web captions on average across the 38 evaluation tasks proposed by DataComp. Refer to the DataComp leaderboard (https://www.datacomp.ai/leaderboard.html) for the top baselines uncovered in our work.
Languages
Primarily English.
Dataset Structure
Data Instances
Each instance maps a unique image identifier from DataComp to the corresponding BLIP2 caption generated with temperature 0.75.
Data Fields
uid: SHA256 hash of image, provided as metadata by the DataComp team.
blip2-cap: corresponding caption generated by BLIP2.
Data Splits
Data was not split. The dataset is intended for pre-training multimodal models.
Dataset Creation
Curation Rationale
Web-crawled image-text data can contain a lot of noise, i.e. the caption may not reflect the content of the respective image. Filtering out noisy web data, however, can hurt the diversity of the training set. To address both of these issues, we use image captioning models to increase the number of useful training samples from the initial pool, by ensuring the captions are more relevant to the images. Our work systematically explores the effectiveness of using these synthetic captions to replace or complement the raw text data, in the context of CLIP pre-training.
Source Data
Initial Data Collection and Normalization
The original 1.28M image-text pairs were collected by the DataComp team from Common Crawl. Minimal filtering was performed on the initial data pool (face blurring, NSFW removal, train-test deduplication). We then replaced the original web-crawled captions with synthetic captions generated by BLIP2.
Who are the source language producers?
Common Crawl is the source for images. BLIP2 is the source of the text data.
Annotations
Annotation process
The dataset was built in a fully automated process: captions are generated by the BLIP2 captioning model.
Who are the annotators?
No human annotators are involved.
Personal and Sensitive Information
The images, which we inherit from the DataComp benchmark, already underwent face detection and face blurring. While the DataComp team made an attempt to remove NSFW instances, it is possible that such content may still exist (to a small degree) in this dataset. Due to the large scale nature of this dataset, the content has not been manually verified to be completely safe. Therefore, it is strongly recommended that this dataset be used only for research purposes.
Considerations for Using the Data
Social Impact of Dataset
The publication contains some preliminary analyses of the fairness implication of training on this dataset, when evaluating on Fairface.
Discussion of Biases
Refer to the publication for more details.
Other Known Limitations
Refer to the publication for more details.
Additional Information
Citation Information
@article{nguyen2023improving,
title={Improving Multimodal Datasets with Image Captioning},
author={Nguyen, Thao and Gadre, Samir Yitzhak and Ilharco, Gabriel and Oh, Sewoong and Schmidt, Ludwig},
journal={arXiv preprint arXiv:2307.10350},
year={2023}
}
- Downloads last month
- 106