Merge branch 'master' of skozl.com:e4-pattern

4 files changed, 50 insertions, 4 deletions

diff --git a/evaluate.py b/evaluate.py
index ebfc34e..7ce586b 100755
--- a/evaluate.py
+++ b/evaluate.py
@@ -86,7 +86,8 @@ def test_model(gallery_data, probe_data, gallery_label, probe_label, gallery_cam
     else:
         if args.mahalanobis:
             # metric = 'jaccard' is also valid
-            distances = cdist(probe_data, gallery_data, 'jaccard')
+            cov_inv = np.linalg.inv(np.cov(gallery_data.T)).T
+            distances = cdist(probe_data, gallery_data, 'mahalanobis', VI=cov_inv)
         else:
             distances = cdist(probe_data, gallery_data, 'euclidean')
 
diff --git a/report2/fig/baseline.pdf b/report2/fig/baseline.pdf
new file mode 100644
index 0000000..e6a4794
--- /dev/null
+++ b/report2/fig/baseline.pdf
diff --git a/report2/fig/eucranklist.png b/report2/fig/eucranklist.png
new file mode 100644
index 0000000..9daaa9c
--- /dev/null
+++ b/report2/fig/eucranklist.png
diff --git a/report2/paper.md b/report2/paper.md
index f70f77a..14f2f98 100755
--- a/report2/paper.md
+++ b/report2/paper.md
@@ -1,12 +1,57 @@
 # Summary
+In this report we analysed how distance metrics learning affects classification
+accuracy for the dataset CUHK03. The baseline method used for classification is
+Nearest Neighbors based on Euclidean distance. The improved approach we propose 
+mixes Jaccardian and Mahalanobis metrics to obtain a ranklist that takes into 
+account also the reciprocal neighbors. This approach is computationally more 
+complex, since the matrices representing distances are effectively calculated 
+twice. However it is possible to observe a significant accuracy improvement of 
+around 10% for the $@rank1$ case. Accuracy improves overall, especially for
+$@rankn$ cases with low n.
+
+# Formulation of the Addresssed Machine Learning Problem
+
+## CUHK03
+
+The dataset CUHK03 contains 14096 pictures of people captured from two 
+different cameras. The feature vectors used come from passing the
+rescaled images through ResNet50. Each feature vector contains 2048
+features that we use for classification. The pictures represent 1467 different
+people and each of them appears between 9 and 10 times. The separation of
+train_idx, query_idx and gallery_idx allows to perform taining and validation 
+on a training set (train_idx, adequately split between test, train and 
+validation keeping the same number of identities). This prevents overfitting 
+the algorithm to the specific data associated with query_idx and gallery_idx.
+
+## Probelm to solve
+
+The problem to solve is to create a ranklist for each image of the query set
+by finding the nearest neighbor(s) within a gallery set. However gallery images
+with the same label and taken from the same camera as the query image should
+not be considered when forming the ranklist.
+
+## Nearest Neighbor ranklist
+
+Nearest Neighbor aims to find the gallery image whose feature are the closest to
+the ones of a query image, predicting the class of the query image as the same 
+of its nearest neighbor(s). The distance between images can be calculated through
+different distance metrics, however one of the most commonly used is euclidean
+distance, represented as $d=\sqrt{\sum (x-y)^{2}}$.
+
+EXPLAIN KNN BRIEFLY
 
-# Baseline Formulation
 
 # Baseline Evaluation
 
-# Formulation of Suggested Improvement 
+\begin{figure}
+\begin{center}
+\includegraphics[width=17em]{fig/baseline.pdf}
+\caption{Recognition accuracy of baseline Nearest Neighbor @rank k}
+\label{fig:baselineacc}
+\end{center}
+\end{figure}
 
-# Suggested Improvement Evaluation
+# Suggested Improvement
 
 # Conclusion