From dhilvert at gmail.com Sun Jul 6 01:16:45 2008 From: dhilvert at gmail.com (David Hilvert) Date: Sun Jul 6 01:17:11 2008 Subject: [ale] Want to write OpenGL shader code? Message-ID: <20080706031645.079657b2.dhilvert@gmail.com> git now has a very incomplete conversion of CPU code to GPU code (OpenGL shader language). The typical conversion currently transforms something looking like this: #include "foo.h" class bar { foo a; public: void b() { a->c(1); } bar() { } }; Into something like this: #include "foo.h" #define ALE_GPU_BAR_INCLUDE \ ALE_GPU_FOO_INCLUDE,\ "struct bar {",\ " foo a;",\ "};",\ "bar_b(bar _this);" class bar : public gpu::program::library { foo a; public: void b() { a->c(1); } bar() { const char *shader_code[] = { ALE_GPU_BAR_INCLUDE, "bar_b(bar _this) {", " foo_c(_this.a, 1);", "}", NULL }; if (accel::is_gpu()) { gpu_shader = new shader(shader_code); } } void attach_shaders(gpu::program *p) const { p->attach(gpu_shader); a->attach_shaders(p); } }; The details are a bit different in actual code at the moment, but this is roughly the technique used. In particular, the GLSL code very closely mirrors the CPU code, and it is conceivable that some sort of automatic translation (e.g., via a C++-to-C preprocessor) could be used for some code sections. Right now, I am doing conversion manually, which is taking a bit of time. If anyone is interested in contributing to the effort, this would be helpful, and will probably result in a GPU-accelerated program sooner rather than later. (Estimates of the precise benefit of GPU acceleration are difficult, but if the improvement of direct-rendered over software-rendered GLX is any indication, the benefit could be substantial.) The latest code is in the git repository, at http://repo.or.cz/w/Ale.git . To generate the build environment, run './bootstrap', which requires (at least) the following: automake gnulib autoconf autoconf-archive A GPU-enabled build will also require the following headers and libraries: GLUT (GL Utility Toolkit; e.g., freeglut) GLEW (GL Extension Wrangler) Once the new code is built, the option --accel=gpu will attempt to use GPU-accelerated code (which will invariably fail at the moment). This first checks for OpenGL features that it thinks are necessary, which should give some idea of whether the card and driver combination in use are adequate. The list of features currently checked includes: GL_ARB_fragment_shader GL_ARB_texture_float GL_ARB_texture_rectangle GL_EXT_framebuffer_object To disable GPU acceleration, --accel=none can be used. By default, the wrapper script checks available GL features and enables or disables acceleration accordingly. From sdfwyhn at asrwev.ewqasdd.todayisp.com Wed Jul 9 11:46:15 2008 From: sdfwyhn at asrwev.ewqasdd.todayisp.com (=?GB2312?B?xbfR9NCmw8A=?=) Date: Wed Jul 9 11:46:28 2008 Subject: [ale] Chinese PE rattan outdoor furniture Message-ID: [1]Home [2]Coverall series [3]HOT [4]Leisure series [5]Color board series [6]Factory scene [7]Contact us we recommend our Chinese hand-made PE rattan outdoor furniture to you, which are made in Guangzhou, China. The exquisitely made and moderately priced furniture will be interest of you. our products have been widely sold to various markets abroad. We make the great importance to production quality and try to be perfect in every process from materials chosen, handmade weaving to packing. We enjoy good reputation and have good partnership with most customers from most of regions and countries. The quality of our products receives their praise. We believe that also demand for it at your end. In order to give you a general idea of the scope of our business, now we enclose our website for your reference. Please study it and let us know your comments, so that we can discuss the business possibilities. If you will give us your specific inquiry, that will be most welcome and reply you as soon as possible.(We cannot read them if you reply us directly, please sent emalis to our email address in our website) Warmly welcome to visit our factory. We will provide warm and thoughtful all-round service. We sincerely hope that we can build mutual trust and win-win cooperation in the long term. And we believe that our cooperation will have a bright prospect in the future. We hope to hear from you soon. Contact us£ºMary Liu Guangzhou QianDu Co.,Ltd Tel£º0086-20-85928531 Fax£º0086-20-83692867 Add£ºJiangxia Industrial Park, Baiyun district, Guangzhou, China,510510 [b_8A5BF845CEFF030015FEB7768E9C267D.jpg] If you can not refer to the link and the photos above,pls look at this website£ºhttp://hi.baidu.com/frhub/blog/item/4e3c6f13506754d5f6039e73. html References 1. http://hi.baidu.com/frhub/blog/item/fd9cb40f8ff2e12f6059f386.html 2. http://hi.baidu.com/frhub/blog/item/d1033039d50d0cf63a87ce80.html 3. http://hi.baidu.com/frhub/blog/item/4e3c6f13506754d5f6039e73.html 4. http://hi.baidu.com/frhub/blog/item/d22a03cd97c9a8540eb34573.html 5. http://hi.baidu.com/frhub/blog/item/556735a275b507a8caefd07c.html 6. http://hi.baidu.com/frhub/blog/item/2c3f930936749aaa2fddd47d.html 7. http://hi.baidu.com/frhub/blog/item/661bdcdb55bf6463d1164e7d.html From christian.ridderstrom at gmail.com Thu Jul 10 04:15:08 2008 From: christian.ridderstrom at gmail.com (=?ISO-8859-1?Q?Christian_Ridderstr=F6m?=) Date: Thu Jul 10 04:16:20 2008 Subject: [ale] ALE and photos of whiteboards Message-ID: Hi, I wonder if anyone on this list has used ALE to improve the quality of photos of whiteboards by mergin several frames? I'm interested in tips on the kind of options I should give to ALE in this case. Caveat: I'm a complete newbie to ALE, and I'm not even sure ALE is a suitable tool. Best regards, Christian -- Christian Ridderstr?m, +46-8-768 39 44 http://www.md.kth.se/~chr From dhilvert at gmail.com Wed Jul 16 07:51:00 2008 From: dhilvert at gmail.com (David Hilvert) Date: Wed Jul 16 07:51:14 2008 Subject: [ale] --visp (and other things) without --ex Message-ID: <20080716095100.8aba9908.dhilvert@gmail.com> git HEAD now has a tracking option --track-primary, which may be usable as a substitute for --ex in --visp (and other) runs. A reasonable configuration for 352x280 video seems to be: --track-primary --md 20 --multi=llocal --achain last:triangle:2 --dchain last:triangle:2 Larger frames could probably use larger --md values. --track-primary currently works by reducing positional error for a subset of points drawn from the multi-alignment pass. Since it relies on non-trivial multi-alignment, --multi and --md settings are important. The file d2/trans_multi.h includes various conditionally-compiled code that may be worth experimenting with in the case that the current --track-primary algorithm is not adequate. Note that there appears to be a memory leak somewhere at the moment; I hadn't noticed this earlier, as I hadn't used --visp for a while. Note also that the command-line option processor is buggy, and may sometimes reject valid input.