mychina75

Waiting for the caffe solution too...

Hi, Still can't access the data… Please help, thanks~

I tested the solution. loss computing may has errors. loss drops fast first, and after few hundreds of iters, loss will getting bigger and bigger.

loss can steady reduce now, but the evaluation result of model getting worse... ############## Line 7015: I0904 13:14:38.944118 13639 solver.cpp:546] Test net output #0: detection_eval = 0.0882107 Line 8234: I0904...

@chuanqi305 I trained the model on COCO 80, looks like model getting worse. but loss values normal... never met with this before, so I stopped training early. any clue about...

@zhanglonghao1992 no... I can not get better result.. maybe need to change some parameters?

啥时间更新预编译的ncnn-android-vulkan-lib？ 想在手机上跑跑超分的模型，APK里面。 不过好像需要下面的几个库... ncnn-android-vulkan-lib/${ANDROID_ABI}/libglslang.a ncnn-android-vulkan-lib/${ANDROID_ABI}/libOGLCompiler.a ncnn-android-vulkan-lib/${ANDROID_ABI}/libOSDependent.a ncnn-android-vulkan-lib/${ANDROID_ABI}/libSPIRV.a)

多谢大神！

@nihui 帮忙看看是啥问题吧。 在手机上跑了下，512x512的输入 模型load需要2s多，process时间6s多。输出的结果也不太对，保存的图是黑的... 如果用原图尺寸的话，3xxx X 2xxx跑一下时间更久，要几十秒 大概的调用是这样的： // ncnn from bitmap ncnn::Mat in = ncnn::Mat::from_android_bitmap_resize(env, bitmap, ncnn::Mat::PIXEL_BGR2RGB,512,512); int c = 3; ncnn::Mat out = ncnn::Mat(512 * scale, 512 *...

这段代码获取的heap_budget是一千多， tilesize就设成了200; if (tilesize == 0) { uint32_t heap_budget = ncnn::get_gpu_device(gpuid)->get_heap_budget(); // more fine-grained tilesize policy here if (model.find(PATHSTR("models-srmd")) != path_t::npos) { if (heap_budget > 2600) tilesize = 400; else...