{ "cells": [ { "cell_type": "markdown", "metadata": { "collapsed": true }, "source": [ "#### Problem definition:\n", "\n", "Given a video at duration: $D$,\n", "\n", "By running a shot-detection algorithm, we get the information of n shots: $v_1, v_2, v_3...v_n$ . (each shot length varies from 2 sec - 5 sec.)\n", "\n", "Based on the computer vision algorithm, we also have the score of each shot: $s_1, s_2, s_3, ...s_n$ and the duration of each shot: $d_1, d_2, .... d_n$.\n", "\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "From the prior research, people have reported the correlation between the playback rate and the understand rate:\n", "\n", "Understand rate is referenced to the ratio of people can well understand the content for certain speed.\n", "\n", "\n", "for silent parts: $u_s = -0.1 * (v_s -1 ) + 1$ \n", "\n", "for non-silent parts: $u_n = - (v_n - 1) + 1$" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "\n", "Our preliminary study shows users prefer a universal constant speed across the video.\n", "\n", "Based on this finding, we maintain a universal $v_s$ and $v_n$ for all the shots.\n", "\n", "Meanwhile, we want to maintain an equal understanding for the silent parts and non-silent parts. \n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "As we speed up the content, the actual score ${s_k}^*$ can be perceived by the user is:\n", "\n", "${s_k}^* = u_s * {s_k}$\n", "\n", "The actual length for each shot is:\n", "\n", "${d_k}^* = \\frac{{d\\_{ks}}}{v_s} + \\frac{{d\\_{kn}}}{v_n}$" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "The final question is:\n", "\n", "How can we maximum the total scores of different shots under a certain time budget?" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "#### Solution description:\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Run a dynamic programming over the different speedup case.\n", "- $v_s$ range: 1x-6x; $v_n$ range: 1x-1.5x. \n", "- speed up step count: 100.\n", "- actual speedup case [$v_s, v_n$]: [1x, 1x], [1.5x, 1.05x], [2x, 1.1x].... [6x, 1.5x]\n", " \n", " \n", "Maintain a hashmap of dynamic map for that 10 speedup case.\n", "\n", "\n", "For any time budget, we iterate different speedup cases and find the optimal solution." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "#### Change to the evaluation:\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "- how should we evaluate the different recall in the current case?\n", "\n", "- what's the ground truth strategy for different time budgets now?" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "collapsed": true }, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "Python 2", "language": "python", "name": "python2" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 2 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython2", "version": "2.7.11" } }, "nbformat": 4, "nbformat_minor": 0 }