Skip to content


Subversion checkout URL

You can clone with
Download ZIP
Browse files

Expand inter-domain text

  • Loading branch information...
1 parent 3b2c849 commit 2b391e7ee6f3a9f5af38cba90566b1cf7b9cdaa0 Felix Stürmer committed
36 thesis/chapters/chapter04/intra_domain_results.tex
@@ -2,13 +2,35 @@ \section{Intra-Domain Results}\label{sec:intra_results}
As described in Section~\ref{sec:results_benchmarking}, the benchmark on a
dataset consisting purely of sketches is meant to remove the edge detection
-preprocessing steps as possible biases. The pipeline configuration used is
-shown in Figure~\ref{fig:pipeline_local_luma_pmean}.
+preprocessing steps as possible biases. The pipeline configurations used are
+the global LUMA+MEAN variant (Figure~\ref{fig:pipeline_global_luma_mean}) and
+the local LUMA+PMEAN pipeline (Figure~\ref{fig:pipeline_local_luma_pmean}). As
+the two graphs in Figure~\ref{fig:results_precision} show, there is a large
+spread of precision values across the different categories. This is in line
+with the observations made by the creators of the benchmark dataset
+\autocite{eitz_how_2012}, that state that the images appear to be of varying
+difficulty for computational classification. Here, for example, the descriptors
+seem to deal with the "pear" or "donut" category well, and fail to categorize
+images from the "cabinet" category to any significant degree.
- \input{illustrations/graphs/pr_l_luma_pmean.tex}
- \caption[Precisions]{
+ \subfloat[Precisions for global LUMA+MEAN]{%
+ %\input{illustrations/graphs/pr_g_luma_mean.tex}
+ \label{fig:results_precision_global_luma_mean}
- \label{tab:results_precision_local_luma_pmean}
+ \quad
+ \subfloat[Precisions for local LUMA+PMEAN]{%
+ \input{illustrations/graphs/pr_l_luma_pmean.tex}
+ \label{fig:results_precision_local_luma_pmean}
+ }
+ \caption[Precision and Recall Results]{
+ The graphs show results of applying the global LUMA+MEAN
+ \subref{fig:results_precision_global_luma_mean} and local LUMA+PMEAN
+ \subref{fig:results_precision_global_luma_mean} pipelines to
+ intra-domain retrieval of sketches. The areas indicate the overall
+ spread of values across all categories while the line plots show
+ results for selected categories.
+ }
+ \label{fig:results_precision}
BIN  thesis/thesis.pdf
Binary file not shown

0 comments on commit 2b391e7

Please sign in to comment.
Something went wrong with that request. Please try again.