The IPOL Initiative: Publishing and Testing Algorithms on Line for Reproducible Research in Image Processing

IPOL: Reviewed publication and public testing of research software

“…The implementation can only depend on a list of allowed essential 6 Style guides since the original Indian Hill Recommended C Style and Coding Standards have been collected and archived by Chris Lott (http:// www.maultech.com/chrislott/resources/cstyle/). 7 These priorities are close to the "six features of useful machine learning software" as defined by the JMLR and MLOSS projects [16], [17], [12]: [useful] software is usable, documented, robust, it has well-defined interfaces, it uses existing interfaces and standards, it has testing routines. 8 libraries chosen by the editorial board for their quality, portability and stability.…”

“…But with the current guidelines, we only focus on our priority: correct, readable and usable programs. 7 These guidelines may seem obvious for a seasoned programmer. Compared to the stringent rules enforced in some branches of the software industry, these requirements are minimal.…”

Using Lepton for documenting source code : a guided example in computer vision

“…if (argc != 6+1) { printf("This program expects 6 argument\n"); printf("Usage : linear original.pgm solution.pgm decomposition.pgm threshold flevel llevel type\n"); printf(" original.pgm : source image, square, power of two, PGM format\n"); printf(" solution.pgm : source image, square, power of two, PGM format\n"); printf(" decomposition.pgm : source image, square, power of two, PGM format\n"); printf(" threshold : floating-point number\n"); printf(" llevel : int, lower level\n"); printf(" type : int, type of normalization L1 (1), L2 (2) or Linf (3)\n"); return EXIT_FAILURE; } char* init = argv [1]; /* source image */ char* fin = argv [2]; /* output image */ char* inter = argv [3]; /* decomposition image */ float p = atof(argv [4]); /* Threshold */ int e0 = atoi(argv [5]); /* lower level */ int resn = atoi(argv [6]);/* type of normalization */ Similarly, the logic for reading the input image and writing the results to disk is shared between the five programs. At this point, we discovered that the methods could only be applied to square images of size power of two.…”

“…However, there is growing realization that our current procedures for collaboration and publication are not optimal. In many fields of study such as econometrics [2], biostatistics [7] and image analysis [6], many published scientific results cannot be reproduced independently because essential elements of the methods are not available or insufficiently documented.…”

Service Level Management for Executable Papers

“…Propriety software used by experiments is also subject to strict licensing rules. The papers published in the grand challenge workshop propose various solutions to realize the executable paper concepts [6][7][8][9]. The papers focuses on the technical details and technology choices but give little attention to the operational aspects associated with the deployment of such a service and what would be the impact on the stakeholders to provide a reliable and scalable service allowing re-execution of published scientific.…”