From da6aa60791e77a8dd0ebec10eda1185cd9fd1d09 Mon Sep 17 00:00:00 2001 From: nunzip Date: Fri, 14 Dec 2018 03:14:37 +0000 Subject: Fix grammar - Make non understandable parts bold --- report/metadata.yaml | 3 +-- report/paper.md | 23 ++++++++++++----------- 2 files changed, 13 insertions(+), 13 deletions(-) diff --git a/report/metadata.yaml b/report/metadata.yaml index 74732a6..3eaadb0 100755 --- a/report/metadata.yaml +++ b/report/metadata.yaml @@ -17,7 +17,6 @@ abstract: | The improved approach evaluated utilises Jaccardian metrics to rearrange the NN ranklist based on reciprocal neighbours. While this approach is more complex and introduces new hyperparameters, significant accuracy improvements are observed - - approximately 10% higher Top-1 identification, and good improvements for Top-$N$ - accuracy with low $N$. + approximately 10% higher Top-1 identification, and good improvements for Top-$N$ accuracy with low $N$. ... diff --git a/report/paper.md b/report/paper.md index 83cde10..8502d9d 100755 --- a/report/paper.md +++ b/report/paper.md @@ -7,11 +7,11 @@ pedestrian images from disjoint cameras by pedestrian detectors. This problem is challenging, as identities captured in photos are subject to various lighting, pose, blur, background and occlusion from various camera views. This report considers features extracted from the CUHK03 dataset, following a 50 layer Residual network -(ResNet50). This paper considers distance metrics techniques which can be used to -perform person re-identification across *disjoint* cameras, using these features. +(ResNet50). Different distance metrics techniques can be used to +perform person re-identification across *disjoint* cameras. Features extracted from Neural Networks such as ResNet-50 are already highly processed -and represent features extracted from the raw data. We therefore expect it to be extremely +and represent **features** extracted from the raw data. We therefore expect it to be extremely hard to further optimise the feature vectors for data separation, but we may be able benefit from alternative neighbour matching algorithms that take into account the relative positions of the nearest neighbours with the probe and each other. @@ -22,7 +22,7 @@ The dataset CUHK03 contains 14096 pictures of people captured from two different cameras. The feature vectors used, extracted from a trained ResNet50 model , contain 2048 features that are used for identification. -The pictures represent 1467 different identities, each of which appears 9 to 10 +The pictures represent 1467 different identities, each of which appears 7 to 10 times. Data is separated in train, query and gallery sets with `train_idx`, `query_idx` and `gallery_idx` respectively, where the training set has been used to develop the ResNet50 model used for feature extraction. This procedure has @@ -120,17 +120,17 @@ The Mahalanobis distance metric was used to create the ranklist as an alternativ $$ d_M(p,g_i) = (p-g_i)^TM(p-g_i). $$ -When performing Mahalanobis with the covariance matrix $M$ generated from the training set, reported accuracy is reduced to **38%** . +When performing Mahalanobis with the covariance matrix $M$ generated from the training set, reported accuracy is reduced to 38%. We also attempted to perform the same Mahalanobis metric on a reduced PCA featureset. PCA performed with the top 100 eigenvectors reduces the feature space while keeping 94% of the data variance. This allowed for significant execution time improvements due to the greatly reduced computation requierments for smaller featurespace, but nevertheless demonstrated no improvements over an euclidean metric. -These results are likely due to the **extremely** low covariance of features in the training set. -This is evident when looking at the Covariance matrix of the training data, and is also +These results are likely due to the extremely low covariance of features in the training set. +This is evident when looking at the covariance matrix of the training data, and is also visible in figure \ref{fig:subspace}. This is likely the result of the feature -transformations performed the the ResNet-50 convolution model the features were extracted from. +transformations performed by the ResNet-50 convolution model from which features were extracted. \begin{figure} \begin{center} @@ -141,7 +141,8 @@ transformations performed the the ResNet-50 convolution model the features were \end{center} \end{figure} -While we did not use Mahalanobis as a primary distance metric, it is possible to use the Mahalanobis metric, together with the next investigated solution involving $k$-reciprocal re-ranking. +While we did not use Mahalanobis as a primary distance metric, it is possible to use the Mahalanobis metric, +together with the next investigated solution involving $k$-reciprocal re-ranking. # Suggested Improvement @@ -255,9 +256,9 @@ The difference between the top $k$ accuracies of the two methods gets smaller as The improved results due to $k$-reciprocal re-ranking may be explained by considering that re-ranks based on second order neighbours, that is, the neighbours of the neighbours. For neighbours which display identifiable features, such as a backpack or binder that is -not visible in the query but visible in a close neighbour, the reranking algorithm is able to infer that the strong relationship based on this newly introduced +not visible in the query but visible in a close neighbour, the reranking algorithm is able to infer **that the strong relationship based on this newly introduced feature such as a backpack or folder by the neighbour, uniquely identify other identities in the gallery with the same feature, and moving them up the rankinlist -as a result despite the identifiable feature being hidden in the query. An example of this can be seen in figure \ref{fig:rerank}. +as a result despite the identifiable feature being hidden in the query**. An example of this can be seen in figure \ref{fig:rerank}. \begin{figure} \begin{center} -- cgit v1.2.3