目录
- CONTRIBUTING.mdRename CONTRIBUTING to CONTRIBUTING.md9年前
- DISCLAIMERAdd existing files9年前
- LICENSEAdd existing files9年前
- README.mdUpdate README.md9年前
- aircraft_classes.txtAdd existing files9年前
- aircraft_urls.txt.7zRemove original text files and add their compressed counter parts. Also update README9年前
- bird_classes.txtAdd existing files9年前
- bird_urls1.txt.7zRemove original text files and add their compressed counter parts. Also update README9年前
- bird_urls2.txt.7zRemove original text files and add their compressed counter parts. Also update README9年前
- birdsnap_classes.txtAdd existing files9年前
- cub_classes.txtAdd existing files9年前
- dog_classes.txtAdd existing files9年前
- dog_urls.txt.7zRemove original text files and add their compressed counter parts. Also update README9年前
- fgvc_classes.txtAdd existing files9年前
- flickr_aircraft.txtAdd existing files9年前
- flickr_bird.txtAdd existing files9年前
- flickr_lepidoptera.txtAdd existing files9年前
- lepidoptera_classes.txtAdd existing files9年前
- lepidoptera_urls.txt.7zRemove original text files and add their compressed counter parts. Also update README9年前
- sdog_classes.txtAdd existing files9年前
- sdogs_active_learning_iteration1.txtAdd existing files9年前
- sdogs_active_learning_iteration2.txtAdd existing files9年前
邀请码
版权所有：中国计算机学会技术支持：开源发展技术委员会
京ICP备13000930号-9
京公网安备 11010802032778号
Goldfinch: GOogLe image-search Dataset for FINe grained CHallenges
Goldfinch is a dataset for fine-grained recognition challenges. It contains a list of bird, butterfly, aircraft, and dog categories with relevant Google image search and Flickr search URLs. In addition, it also includes a set of active learning annotations on dog categories. Goldfinch was published along with our ECCV’16 paper The Unreasonable Effectiveness of Noisy Data for Fine-Grained Recognition.
Here we list the files included in this release and their formats. Where possible, we have mapped category names to freebase ids as a form of canonicalization. Questions can be directed to jkrause@google.com and howardzhou@google.com.
Class files
cub_classes.txt:
For each category in the CUB dataset:
E.g.
birdsnap_classes.txt:
For each category in the Birdsnap dadataset:
E.g.
fgvc_classes.txt:
For each category in the FGVC-Aircraft dataset:
Note: no freebase id is used for aircraft since not all aircraft in FGVC correspond to a freebase entity. E.g.
sdog_classes.txt:
For each category in the Stanford Dogs dataset:
E.g.
bird_classes.txt:
For each category in L-Bird:
E.g.
lepidoptera_classes.txt:
For each category in L-Butterfly:
E.g.
dog_classes.txt:
For each category in L-Dog:
E.g.
aircraft_classes.txt:
For each category in L-Aircraft:
E.g.
Image URLs
Compressed version of URL text files. Github has a 100MB file size limit. As a result, we had to split bird_urls into 2 files: bird_urls1, bird_urls2 and compress them separately.
aircraft_urls.txt.7z
bird_urls1.txt.7z
bird_urls2.txt.7z
dog_urls.txt.7z
lepidoptera_urls.txt.7z
For each image search image:
Note that if more than one url is given, that indicates the same image is hosted at more than one url, which we provide for convenience in case one of the urls goes down. Where possible, we have provided links to full resolution images, but when not possible a link to a thumbnail is provided.
E.g.
flickr_{aircraft,bird,lepidoptera}.txt:
For each image in the Flickr testing set:
E.g.
Active learning files
sdogs_active_learning_iteration{1,2}.txt:
Images mined from Yahoo Flickr Creative Commons 100M dataset. Format of each line, with values delimited by commas:
E.g.
Questions can be directed to jkrause@google.com and howardzhou@google.com.