{ "access": { "embargo": { "active": false, "reason": null }, "files": "restricted", "record": "public", "status": "restricted" }, "created": "2018-05-05T07:10:26.359900+00:00", "custom_fields": { "meeting:meeting": { "acronym": "CLEF", "dates": "8-11 September 2015", "place": "Toulouse, France", "title": "Conference and Labs of the Evaluation Forum", "url": "http://clef2015.clef-initiative.eu/" } }, "deletion_status": { "is_deleted": false, "status": "P" }, "files": { "enabled": true }, "id": "1038547", "is_draft": false, "is_published": true, "links": { "access": "https://zenodo.org/api/records/1038547/access", "access_links": "https://zenodo.org/api/records/1038547/access/links", "access_request": "https://zenodo.org/api/records/1038547/access/request", "access_users": "https://zenodo.org/api/records/1038547/access/users", "archive": "https://zenodo.org/api/records/1038547/files-archive", "archive_media": "https://zenodo.org/api/records/1038547/media-files-archive", "communities": "https://zenodo.org/api/records/1038547/communities", "communities-suggestions": "https://zenodo.org/api/records/1038547/communities-suggestions", "doi": "https://doi.org/10.5281/zenodo.1038547", "draft": "https://zenodo.org/api/records/1038547/draft", "files": "https://zenodo.org/api/records/1038547/files", "latest": "https://zenodo.org/api/records/1038547/versions/latest", "latest_html": "https://zenodo.org/records/1038547/latest", "media_files": "https://zenodo.org/api/records/1038547/media-files", "parent": "https://zenodo.org/api/records/1038546", "parent_doi": "https://zenodo.org/doi/10.5281/zenodo.1038546", "parent_html": "https://zenodo.org/records/1038546", "requests": "https://zenodo.org/api/records/1038547/requests", "reserve_doi": "https://zenodo.org/api/records/1038547/draft/pids/doi", "self": "https://zenodo.org/api/records/1038547", "self_doi": "https://zenodo.org/doi/10.5281/zenodo.1038547", "self_html": "https://zenodo.org/records/1038547", "self_iiif_manifest": "https://zenodo.org/api/iiif/record:1038547/manifest", "self_iiif_sequence": "https://zenodo.org/api/iiif/record:1038547/sequence/default", "versions": "https://zenodo.org/api/records/1038547/versions" }, "media_files": { "enabled": false }, "metadata": { "creators": [ { "affiliations": [ { "name": "University of Surrey" } ], "person_or_org": { "family_name": "Gilbert", "given_name": "Andrew", "name": "Gilbert, Andrew", "type": "personal" } }, { "affiliations": [ { "name": "University of Cagliari" } ], "person_or_org": { "family_name": "Piras", "given_name": "Luca", "name": "Piras, Luca", "type": "personal" } }, { "affiliations": [ { "name": "University of Sheffield" } ], "person_or_org": { "family_name": "Wang", "given_name": "Josiah", "name": "Wang, Josiah", "type": "personal" } }, { "affiliations": [ { "name": "University of Surrey" } ], "person_or_org": { "family_name": "Yan", "given_name": "Fei", "name": "Yan, Fei", "type": "personal" } }, { "affiliations": [ { "name": "Ecole Centrale de Lyon" } ], "person_or_org": { "family_name": "Dellandrea", "given_name": "Emmanuel", "name": "Dellandrea, Emmanuel", "type": "personal" } }, { "affiliations": [ { "name": "University of Sheffield" } ], "person_or_org": { "family_name": "Gaizauskas", "given_name": "Robert", "name": "Gaizauskas, Robert", "type": "personal" } }, { "affiliations": [ { "name": "Universitat Politecnica de Valencia" } ], "person_or_org": { "family_name": "Villegas", "given_name": "Mauricio", "name": "Villegas, Mauricio", "type": "personal" } }, { "affiliations": [ { "name": "University of Surrey" } ], "person_or_org": { "family_name": "Mikolajczyk", "given_name": "Krystian", "name": "Mikolajczyk, Krystian", "type": "personal" } } ], "description": "
This document describes the WEBUPV dataset compiled for the ImageCLEF 2015
\nScalable Concept Image Annotation challenge. The data mentioned here indicates
\nwhat is ready for download. However, upon request or depending on feedback
\nfrom the participants, additional data may be released.
The following is the directory structure of the collection, and bellow there
\nis a brief description of what each compressed file contains. The
\ncorresponding MD5 checksums of the files shown (for verifying a correct
\ndownload) can be found in md5sums.txt.
\nDirectory structure
\n-------------------
.
\n|
\n|--- README.txt
\n|--- md5sums.txt
\n|--- webupv15_data_lists.zip
\n|
\n|--- annotations/imageclef2015.dev.bbox.*.gz
\n|--- annotations/imageclef2015.dev.textdesc.*.gz
\n|--- annotations/imageclef2015.subtask2cleantrack.*.input_bbox.*.gz
\n|--- annotations/imageclef2015.subtask2cleantrack.dev.textdesc.*.gz
\n|--- annotations/imageclef2015.dev.bbox.supplement.*.gz
\n|--- annotations/imageclef2015.hierarchy.*.gz
\n|--- annotations/imageclef2015.concept_to_parents.*.gz
\n|
\n|--- feats_textual/
\n| |
\n| |--- webupv15_data_textual_pages.zip
\n| |--- webupv15_data_textual.scofeat.gz
\n| |--- webupv15_data_textual.keywords.gz
\n|
\n|--- feats_visual/
\n |
\n |--- webupv15_data_visual_images.zip
\n |--- webupv15_data_visual_gist.dfeat.gz
\n |--- webupv15_data_visual_sift_1000.sfeat.gz
\n |--- webupv15_data_visual_csift_1000.sfeat.gz
\n |--- webupv15_data_visual_rgbsift_1000.sfeat.gz
\n |--- webupv15_data_visual_opponentsift_1000.sfeat.gz
\n |--- webupv15_data_visual_colorhist.sfeat.gz
\n |--- webupv15_data_visual_getlf.sfeat.gz
\n |--- webupv15_data_visual_vgg16-fc8.dfeat.gz
\nContents of files
\n-----------------
* annotations/imageclef2015.dev.bbox.*.gz
\n\nDevelopment set ground truth localised annotations for sub task 1.
\n\n The format for the development set of annotated bounding boxes of
\n the concepts is
<image_ID> <seq> <Concept> <confidence> <xmin> <ymin> <xmax> <ymax>
\n\n The development set contains 1,979 images. The bounding boxes may enclose
\n single instances (a single tree) or grouped instances (e.g. group of trees),
\n depending on the context. The annotations are not exhaustive: the emphasis
\n is on concepts that are interesting enough to be described in the image,
\n although background objects are also optionally annotated by our annotators
\n in many cases. Also note that a person might not be annotated if the
\n annotator could not decide whether the person is a man/woman/boy/girl.
\n* annotations/imageclef2015.dev.textdesc.*.gz
Development set ground truth textual description annotations of images for
\n sub task 2
The format is:
\n\n<image_ID> \\t <text_description_seq> \\t <textual_description>
\n\n The development set contains 2,000 images with 5 to 51 textual descriptions
\n per image (mean: 9.492, median: 8). Please note that the sentences contain a
\n mix of both American and British English spelling variants (e.g. color vs
\n colour) -- we have decided to retain this variation in the annotations to
\n reflect the challenge of real-world English spelling variants. Basic
\n spell-correction has been performed on the textual descriptions, but we cannot
\n guarantee that they are completely free from spelling or grammatical error.
Changelog for v20150306: We have removed some poor textual descriptions that
\n managed to slip past our quality control check. Please use this latest version
\n (v20150306) for your development needs.
\n* annotations/imageclef2015.subtask2cleantrack.*.input_bbox.*.gz
Input bounding boxes for the clean track of SubTask 2. This is just a selected
\n subset of 500 development images from imageclef2015.dev.bbox above (please refer
\n to above for file format).
\n* annotations/imageclef2015.subtask2cleantrack.dev.textdesc.*.gz
Annotated textual descriptions for 500 development images, to be used to evaluate
\n the content selection ability of the text generation system in the clean track of
\n SubTask 2. The format is the same as the original imageclef2015.dev.textdesc
\n file, except that we further annotated textual terms with their corresponding
\n input bounding boxes, for example [[[dogs|0,4]]] in a textual description refers
\n to the two instances of dogs with the bounding box id 0 and 4 in
\n imageclef2015.subtask2cleantrack.dev.input_bbox.
Note that not all descriptions from the original imageclef2015.dev.textdesc are
\n used in this version, and as such the sequence numbers of the descriptions may not
\n necessarily be contiguous as we retained the sequence numbers from the original
\n file for consistency.
\n* annotations/imageclef2015.dev.bbox.supplement.*.gz
This file contains all bounding boxes from imageclef2015.dev.bbox, and additional
\n bounding boxes for additional 'general level' categories ('building', 'person',
\n 'animal' etc.) not in the 251 concepts list. These special categories are prefixed
\n with an asterisk(*). We hope that you will find these useful for your development
\n purposes.
\n* annotations/imageclef2015.hierarchy.*.gz
The hierarchy structure of the 'general level' categories. File format:
\n\n*category \\t *parent-category \\t definition.
\n\nFor example, *mammal is the child of *animal. '#' represents the root node.
\n\n
\n* annotations/imageclef2015.concept_to_parents.*.gz
List of 'general level' category parent(s) for each 251 concept. A concept may
\n have multiple parents (separated by commas). File format:
category \\t *parent1,*parent2
\n\n
\n* annotations/imageclef2015.test.groundtruth.zip
\n\n
Ground truth for the test.
\n\n\n\n
\n* webupv15_data_lists.zip
-> data_iids.txt : IDs of the images (IIDs) in the dataset.
\n\n-> data_rids.txt : IDs of the webpages (RIDs) in the dataset.
\n\n -> data_*urls.txt : The original URLs from where the images (iurls)
\n and the webpages (rurls) were downloaded. Each line in the file
\n corresponds to an image, starting with the IID and is followed
\n by one or more URLs.
-> data_rimgsrc.txt : The URLs of the images as referenced in each
\n of the webpages. Each line of the file is of the form: IID RID
\n URL1 [URL2 ...]. This information is necessary to locate the
\n images within the webpages and it can also be useful as a
\n textual feature.
\n* feats_textual/webupv15_data_textual_pages.zip
Contains all of the webpages which referenced the images in the
\n dataset set after being converted to valid xml. In total there are
\n 515754 files, since each image can appear in more than one page, and
\n there can be several versions of same page which differ by the
\n method of conversion to xml. To avoid having too many files in a
\n single directory (which is an issue for some types of partitions),
\n the files are found in subdirectories named using the first two
\n characters of the RID, thus the paths of the files after extraction
\n are of the form:
./WEBUPV/pages/{RID:0:2}/{RID}.{CONVM}.xml.gz
\n\n To be able to locate the images withing the webpages, the URLs of the
\n images as referenced are provided in the file data_rimgsrc.txt.
\n* feats_textual/webupv15_data_textual.scofeat.gz
The processed text extracted from the webpages near where the images
\n appeared. Each line corresponds to one image, having the same order
\n as the data_iids.txt list. The lines start with the image ID,
\n followed by the number of extracted unique words and the
\n corresponding word-score pairs. The scores were derived taking into
\n account 1) the term frequency (TF), 2) the document object model
\n (DOM) attributes, and 3) the word distance to the image. The scores
\n are all integers and for each image the sum of scores is always
\n <=100000 (i.e. it is normalized).
\n* feats_textual/webupv15_data_textual.keywords.gz
The words used to find the images when querying image search
\n engines. Each line corresponds to an image (in the same order as in
\n data_iids.txt). The lines are composed of triplets:
[keyword] [rank] [search_engine]
\n\n where [keyword] is the word used to find the image, [rank] is the
\n position given to the image in the query, and [search_engine] is a
\n single character indicating in which search engine it was found
\n ('g':google, 'b':bing, 'y':yahoo).
\n* feats_visual/webupv15_data_visual_images.zip
Contains thumbnails (maximum 640 pixels of either width or height)
\n of the images in jpeg format. To avoid having too many files in a
\n single directory (which is an issue for some types of partitions),
\n the files are found in subdirectories named using the first two
\n characters of the IID, thus the paths of the files after extraction
\n are of the form:
./WEBUPV/images/{IID:0:2}/{IID}.jpg
\n\n
\n* feats_visual/webupv15_*.{s|d}feat.gz
The visual features in a simple ASCII text format either in sparse
\n (*.sfeat.gz files) or dense (*.dfeat.gz files). The first
\n line of the file indicates the number of vectors (N) and the
\n dimensionality (DIMS). Then each line corresponds to one vector.
\n For the dense features each line has exactly DIMS values separated
\n by spaces, i.e., the format is:
N DIMS
\n Val(1,1) Val(1,2) ... Val(1,DIMS)
\n Val(2,1) Val(1,2) ... Val(2,DIMS)
\n ...
\n Val(N,1) Val(N,2) ... Val(N,DIMS)
For the sparse features, each line starts with the number of non-zero
\n elements and is followed by dimension-value pairs, being the first
\n dimension 0, i.e., the format is:
N DIMS
\n nz1 Dim(1,1) Val(1,1) ... Dim(1,nz1) Val(1,nz1)
\n nz2 Dim(2,1) Val(2,1) ... Dim(2,nz2) Val(2,nz2)
\n ...
\n nzN Dim(N,1) Val(N,1) ... Dim(N,nzN) Val(N,nzN)
The order of the features is the same as in the list data_iids.txt.
\n\n The procedure to extract the SIFT based features in this
\n subdirectory was conducted as follows. Using the ImageMagick
\n software, the images were first rescaled to having a maximum of 240
\n pixels, of both width and height, while preserving the original
\n aspect ratio, employing the command:
convert {IMGIN}.jpg -resize '240>x240>' {IMGOUT}.jpg
\n\n Then the SIFT features where extracted using the ColorDescriptor
\n software from Koen van de Sande
\n (http://koen.me/research/colordescriptors). As configuration we
\n used, 'densesampling' detector with default parameters, and a hard
\n assignment codebook using a spatial pyramid as
\n 'pyramid-1x1-2x2'. The number in the file name indicates the size of
\n the codebook. All of the vectors of the spatial pyramid are given in
\n the same line, thus keeping only the first 1/5th of the dimensions
\n would be like not using the spatial pyramid. The codebook was
\n generated using 1.25 million randomly selected features and the
\n k-means algorithm. The GIST features were extracted using the
\n LabelMe Toolbox. The images where first resized to 256x256 ignoring
\n original aspect ratio, using 5 scales, 6 orientations and 4
\n blocks. The other features colorhist and getlf, are both color
\n histogram based extracted using our own implementation.
\n* webupv15_data_visual_vgg16-relu7.dfeat.gz
Contains the 4096 dimensional activations of the relu7 layer of Oxford
\n VGG’s 16-layer CNN model, extracted using the Berkeley Caffe library.
\n More details can be found at https://github.com/BVLC/caffe/wiki/Model-Zoo.
\n* webupv15_data_visual_vgg16-fc8.dfeat.gz
Contains the 1000 dimensional activations of the fc8 layer of Oxford
\n VGG’s 16-layer CNN model, extracted using the Berkeley Caffe library.
\n More details can be found at https://github.com/BVLC/caffe/wiki/Model-Zoo.
\n The 1000 dimensions correspond to the 1000 categories in
\n webupv15_data_visual_vgg16-fc8_categories.lst: the higher the value
\n in the 1000d feature, the more likely the corresponding synset appears
\n in the image.
\nContact
\n-------
For further questions, please contact:
\n Mauricio Villegas <mauvilsa@upv.es>
\n Fei Yan <f.yan@surrey.ac.uk>
", "publication_date": "2015-05-07", "publisher": "Zenodo", "related_identifiers": [ { "identifier": "http://ceur-ws.org/Vol-1391/inv-pap6-CR.pdf", "relation_type": { "id": "issupplementto", "title": { "de": "Erg\u00e4nzt", "en": "Is supplement to" } }, "scheme": "url" }, { "identifier": "http://imageclef.org/2015/annotation", "relation_type": { "id": "issupplementto", "title": { "de": "Erg\u00e4nzt", "en": "Is supplement to" } }, "scheme": "url" } ], "resource_type": { "id": "dataset", "title": { "de": "Datensatz", "en": "Dataset" } }, "title": "2015 ImageCLEF WEBUPV Collection" }, "parent": { "access": { "owned_by": { "user": 19451 }, "settings": { "accept_conditions_text": "
This dataset is available under a Creative Commons Attribution-
\nNonCommercial-ShareAlike 3.0 Unported License. Before downloading
\nthe data, please read and accept the Creative Commons License and
\nthe following usage agreement:
Data Usage Agreement ImageCLEF 2012/2013/2014/2015/2016 WEBUPV Image
\nAnnotation Datasets
By downloading the "Dataset", you (the "Researcher") agrees to the
\nfollowing terms.
* The Researcher will only use the Dataset for non-commercial
\nresearch and/or educational purposes.
* The Researcher will cite one of the following papers in any
\npublication that makes use of the Dataset.
Gilbert, A., Piras, L., Wang, J., Yan, F., Ramisa, A.,
\n Dellandrea, E., Gaizauskas, R., Villegas, M., Mikolajczyk, K.:
\n Overview of the ImageCLEF 2016 scalable concept image
\n annotation task. In: CLEF2016 Working Notes, CEUR Workshop
\n Proceedings, CEUR-WS.org, Évora, Portugal, 5–8 September 2016
Gilbert, A., Piras, L., Wang, J., Yan, F., Dellandrea, E.,
\n Gaizauskas, R., Villegas, M., Mikolajczyk, K.: Overview of the
\n ImageCLEF 2015 Scalable Image Annotation, Localization and
\n Sentence Generation task. In: CLEF2015 Working Notes. CEUR
\n Workshop Proceedings, CEUR-WS.org, Toulouse, France (September
\n 8-11 2015)
Villegas, M., Paredes, R.: Overview of the ImageCLEF 2014
\n Scalable Concept Image Annotation Task. In: CLEF2014 Working
\n Notes. CEUR Workshop Proceedings, vol. 1180, pp. 308–328.
\n CEUR-WS.org, Sheffield, UK (September 15-18 2014)
Villegas, M., Paredes, R., Thomee, B.: Overview of the ImageCLEF
\n 2013 Scalable Concept Image Annotation Subtask. In: CLEF 2013
\n Evaluation Labs and Workshop, Online Working Notes. Valencia,
\n Spain (September 23-26 2013)
* The Researcher may provide research associates and colleagues a
\ncopy of the Dataset provided that they also agree to this Data
\nUsage Agreement.
* The Researcher will assume all responsibility against any claims
\narising from Researcher's use of the Dataset.