This is the homepage for experiments on
"The Berkeley Segmentation Dataset and Benchmark"
[URL: here]
Explanations
This webpage is an attempt to use (and promote the use) of this very interesting material.
This page is available from our pages on our TPAMI and CVPR'04 papers (see these pages for
the binaries used to obtain the results below).
Results will be added on a (hopefully) regular basis.
Experimental setting
Each image of the BSDB was used as is, without preprocessing. For each
image, we provide:
- the image used (as from the BSDB),
- SRM's output,
- a human segmentation from the DB website,
- results of our biased grouping (bias is shown in the same way as for Yu+Shi's
well known papers).
For each image, we add the number of the subgroup to which it belongs in the BSDB.
If you click on the numbers, you will be redirected to the image's page on the BSDB.
Due to space considerations, we do not provide the source .tif image processed: each was
obtained from the BSDB by transforming the .jpg file with xv.
Images and experiments
Desert-2 [ N° 1-25 ]
|
|
|
|
BSDB Image |
SRM |
Human |
Biased grouping |
Waterfall [ N° 1-25 ]
|
|
|
|
BSDB Image |
SRM |
Human |
Biased grouping |
Garden [ N° 26-50 ]
|
|
|
|
BSDB Image |
SRM |
Human |
Biased grouping |
Island [ N° 26-50 ]
|
|
|
|
BSDB Image |
SRM |
Human |
Biased grouping |
Cave [ N° 51-75 ]
|
|
|
|
|
BSDB Image |
SRM |
Human |
Biased grouping (a) |
Biased grouping (b) |
Roc [ N° 76-100 ]
[note: due to the resemblance between SRM and Human, no grouping with bias result is shown]
Storks [ N° 101-125 ]
[note: due to the resemblance between SRM and Human, no grouping with bias result is shown]
Horses-1 [ N° 126-150 ]
|
|
|
|
BSDB Image |
SRM |
Human |
Biased grouping |
Desert [ N° 151-175 ]
|
|
|
|
BSDB Image |
SRM |
Human |
Biased grouping |
Woman-1 [ N° 151-175 ]
|
|
|
|
BSDB Image |
SRM |
Human |
Biased grouping |
Land [ N° 176-200 ]
|
|
|
|
BSDB Image |
SRM |
Human |
Biased grouping |
Feedback
Your feedback is much appreciated. Your can reach us at rnock@martinique.univ-ag.fr,
or at Nielsen@csl.sony.co.jp.