Hi i am tring to build image classification using openimag hadoop tools. So far i have done this
1) Used Sequance File Generater
2) Used Hadoop Local Feature Detector
3) Used HadoopFastKmean to Cluster the Descriptors
4) Using HaddopClusterQuantizer
What will be the next step for Training. And Predicting. I know i have to do Visual histogram and then train either svm or decesion tree.
Please guide me what should i do Next after Step 4.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Take a look at http://www.openimaj.org/tutorial/classification101.html. That should give you some hints to get started. We haven't written a tool to explicitly do the BoVW step (i.e. compress the output of the HadoopClusterQuantiser into a single [potentially sparse] vector) with hadoop, but that should be easy. Whether you need to use hadoop to do the SVM step will depend on how many images you have - the liblinear annotator coupled with the homogeneous kernel map is fairly scalable on a single machine. If you need to scale up, then take a look at the hadoop SVM implementations in Apache Storm and Apache Hadoop.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thankyou for reply, I have little experience in image analysis and most of time i've used matlab.What output formate HadoopClusterQuantiser generate. How do i convert HadoopClusterQuantiser output to single vector? My output file is around 35MB. Most likely i donmt need haddop after step 4.
Last edit: waqas 2015-02-19
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi i am tring to build image classification using openimag hadoop tools. So far i have done this
1) Used Sequance File Generater
2) Used Hadoop Local Feature Detector
3) Used HadoopFastKmean to Cluster the Descriptors
4) Using HaddopClusterQuantizer
What will be the next step for Training. And Predicting. I know i have to do Visual histogram and then train either svm or decesion tree.
Please guide me what should i do Next after Step 4.
Take a look at http://www.openimaj.org/tutorial/classification101.html. That should give you some hints to get started. We haven't written a tool to explicitly do the BoVW step (i.e. compress the output of the HadoopClusterQuantiser into a single [potentially sparse] vector) with hadoop, but that should be easy. Whether you need to use hadoop to do the SVM step will depend on how many images you have - the liblinear annotator coupled with the homogeneous kernel map is fairly scalable on a single machine. If you need to scale up, then take a look at the hadoop SVM implementations in Apache Storm and Apache Hadoop.
Thankyou for reply, I have little experience in image analysis and most of time i've used matlab.What output formate HadoopClusterQuantiser generate. How do i convert HadoopClusterQuantiser output to single vector? My output file is around 35MB. Most likely i donmt need haddop after step 4.
Last edit: waqas 2015-02-19