From 99deea0b69ccd1d7ad8ef889e6da052beda37fef Mon Sep 17 00:00:00 2001 From: nunzip Date: Thu, 13 Dec 2018 15:41:05 +0000 Subject: Minor layout adjustments --- report/paper.md | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-) diff --git a/report/paper.md b/report/paper.md index 79d4183..79612c3 100755 --- a/report/paper.md +++ b/report/paper.md @@ -16,7 +16,7 @@ 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 seperated 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 @@ -41,9 +41,6 @@ distance: $$ \textrm{NN}(x) = \operatorname*{argmin}_{i\in[m]} \|x-x_i\| $$ -Alternative distance metrics exist such as jaccardian and mahalanobis, which can -be used as an alternative to euclidiean distance. - # Baseline Evaluation To evaluate improvements brought by alternative distance learning metrics a baseline @@ -105,21 +102,22 @@ We find that for the query and gallery set clustering does not seem to improve i \end{center} \end{figure} -# Suggested Improvement - ## Mahalanobis Distance We were not able to achieve significant improvements using mahalanobis for original distance ranking compared to square euclidiaen metrics. -The mahalanobis distance metric was used to create the ranklist as an alternative to euclidean distance. -When performing mahalanobis with the training set as the covariance matrix, reported accuracy is reduced to **38%** . +The mahalanobis distance metric was used to create the ranklist as an alternative to euclidean distance: + +$$ d_M(p,g_i) = (p-g_i)^TM(p-g_i). $$ + +When performing mahalanobis with the covariance matrix $M$ generated from the training set, reported accuracy is reduced to **38%** . We also attempted to perform the same mahalanobis metric on a reduced PCA featureset. 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 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. +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. \begin{figure} \begin{center} @@ -130,6 +128,8 @@ These results are likely due to the **extremely** low covariance of features in \end{center} \end{figure} +# Suggested Improvement + ## $k$-reciprocal Re-ranking Formulation The approach addressed to improve the identification performance is based on @@ -167,11 +167,11 @@ e\textsuperscript{\textit{-d(p,g\textsubscript{i})}}, & \text{if}\ \textit{g\tex Through this transformation it is possible to reformulate the distance obtained through Jaccardian metric as: -$$ d_J(p,g_i)=1-\frac{\sum\limits_{j=1}^N min(V_{p,g_j},V_{g_i,g_j})}{\sum\limits_{j=1}^N max(V_{p,g_j},V_{g_i,g_j})} $$ +$$ d_J(p,g_i)=1-\frac{\sum\limits_{j=1}^N min(V_{p,g_j},V_{g_i,g_j})}{\sum\limits_{j=1}^N max(V_{p,g_j},V_{g_i,g_j})}. $$ It is then possible to perform a local query expansion using the g\textsubscript{i} neighbors of defined as: -$$ V_p=\frac{1}{|N(p,k_2)|}\sum\limits_{g_i\in N(p,k_2)}V_{g_i} $$. +$$ V_p=\frac{1}{|N(p,k_2)|}\sum\limits_{g_i\in N(p,k_2)}V_{g_i}. $$ We refer to $k_2$ since we limit the size of the nighbors to prevent noise from the $k_2$ neighbors. The dimension k of the *$R^*$* set will instead be defined as $k_1$: $R^*(g_i,k_1)$. -- cgit v1.2.3 From 628420dbd4183a91cc976bca0102df6d70204333 Mon Sep 17 00:00:00 2001 From: nunzip Date: Thu, 13 Dec 2018 16:28:40 +0000 Subject: Add examples to README --- README.md | 41 ++++++++++++++++++++++++++++++++++++++++- evaluate.py | 2 +- opt.py | 5 ++--- 3 files changed, 43 insertions(+), 5 deletions(-) diff --git a/README.md b/README.md index 029e1a0..8d3f863 100644 --- a/README.md +++ b/README.md @@ -18,7 +18,7 @@ optional arguments: -b RERANKB, --rerankb RERANKB Parameter k2 for rerank -l RERANKL, --rerankl RERANKL - Parameter lambda fo rerank + Parameter lambda for rerank -n NEIGHBORS, --neighbors NEIGHBORS Use customized ranklist size NEIGHBORS -v, --verbose Use verbose output @@ -36,3 +36,42 @@ optional arguments: -P PCA, --PCA PCA Perform pca with PCA eigenvectors ``` +EXAMPLES for `evaluate.py`: + + EXAMPLE 1: Run euclidean distance with top n + `evaluate.py -e -n 10` or simply `evaluate.py -n 10` + + EXAMPLE 2: Run euclidean distance for the first 10 values of top n and graph them + `evaluate.py -M 10` + + EXAMPLE 3: Run comparison between baseline and rerank for the first 5 values of top n and graph them + `evaluate.py -M 5 -C` + + EXAMPLE 4: Run for kmeans, 10 clusters + `evaluate.py -K 10` + + EXAMPLE 5: Run for mahalanobis, using PCA for top 100 eigenvectors to speed up the calculation + `evaluate.py -m -P 100` + + EXAMPLE 6: Run rerank for customized values of RERANKA, RERANKB and RERANKL + `evaluate.py -r -a 11 -b 3 -l 0.3` + + EXAMPLE 7: Run on the training set with euclidean distance and normalize feature vectors. Draw confusion matrix at the end. + `evaluate.py -t -1 -c` + + EXAMPLE 8: Run for rerank top 10 and save the names of the images that compose the ranklist for the first 5 queries: query.txt, ranklist.txt. + `evaluate.py -r -s 5 -n 10` + + EXAMPLE 9: Display mAP. It is advisable to use high n to obtain an accurate results. + `evaluate.py -A -n 5000` + + EXAMPLE 10: Run euclidean distance specifying a different data folder location + `evaluate.py --data` + +EXAMPLES for `opt.py`: + + EXAMPLE 1: optimize top 1 accuracy for k1, k2, lambda speeding up the process with PCA, top 50 eigenvectors + `opt.py -P 50` + + EXAMPLE 2: optimize mAP for k1, k2, lambda speeding up the process with PCA, top 50 eigenvectors + `opt.py -P 50 -A` diff --git a/evaluate.py b/evaluate.py index a19a7a9..9d41424 100755 --- a/evaluate.py +++ b/evaluate.py @@ -39,7 +39,7 @@ parser.add_argument("-e", "--euclidean", help="Use standard euclidean distance", parser.add_argument("-r", "--rerank", help="Use k-reciprocal rernaking", action='store_true') parser.add_argument("-a", "--reranka", help="Parameter k1 for rerank", type=int, default = 9) parser.add_argument("-b", "--rerankb", help="Parameter k2 for rerank", type=int, default = 3) -parser.add_argument("-l", "--rerankl", help="Parameter lambda fo rerank", type=float, default = 0.3) +parser.add_argument("-l", "--rerankl", help="Parameter lambda for rerank", type=float, default = 0.3) parser.add_argument("-n", "--neighbors", help="Use customized ranklist size NEIGHBORS", type=int, default = 1) parser.add_argument("-v", "--verbose", help="Use verbose output", action='store_true') parser.add_argument("-s", "--showrank", help="Save ranklist pics id in a txt file for first SHOWRANK queries", type=int, default = 0) diff --git a/opt.py b/opt.py index e29495e..873b14d 100755 --- a/opt.py +++ b/opt.py @@ -42,7 +42,7 @@ parser.add_argument("-e", "--euclidean", help="Use standard euclidean distance", parser.add_argument("-r", "--rerank", help="Use k-reciprocal rernaking", action='store_true') parser.add_argument("-a", "--reranka", help="Parameter k1 for rerank", type=int, default = 9) parser.add_argument("-b", "--rerankb", help="Parameter k2 for rerank", type=int, default = 3) -parser.add_argument("-l", "--rerankl", help="Parameter lambda fo rerank", type=float, default = 0.3) +parser.add_argument("-l", "--rerankl", help="Parameter lambda for rerank", type=float, default = 0.3) parser.add_argument("-n", "--neighbors", help="Use customized ranklist size NEIGHBORS", type=int, default = 1) parser.add_argument("-v", "--verbose", help="Use verbose output", action='store_true') parser.add_argument("-s", "--showrank", help="Save ranklist pics id in a txt file for first SHOWRANK queries", type=int, default = 0) @@ -52,7 +52,7 @@ parser.add_argument("-C", "--comparison", help="Compare baseline and improved me parser.add_argument("--data", help="Folder containing data", default='data') parser.add_argument("-K", "--kmean", help="Perform Kmean clustering, KMEAN number of clusters", type=int, default=0) parser.add_argument("-A", "--mAP", help="Display Mean Average Precision", action='store_true') -parser.add_argument("-P", "--PCA", help="Perform pca with PCA eigenvectors", type=int, default=0) +parser.add_argument("-P", "--PCA", help="Perform pca with PCA eigenvectors", type=int, default=50) args = parser.parse_args() @@ -118,7 +118,6 @@ def kopt(camId, filelist, labels, gallery_idx, train_idx, feature_vectors, args) start = np.array([1,1]) if args.mAP: args.neighbors = 10 - args.PCA = 50 args.train = True args.rerank = True args.reranka = 1 -- cgit v1.2.3