Liang Su

Results 7 issues of Liang Su

hello, thanks for your great work! I learned a lot from this repo, and i wonder if there is any model here for segment task, which is a common and...

hello, i saw the update about opencv dnn inference module, but i can not get landmark predict by using the opencv dnn module, does onnx module predict with opencv dnn...

你好,考虑到模型加速的问题,请问下使用libtorch训练得到的pt模型,如何导出onnx?或者如何直接通过tensorRT加速?

你好,看了下unet结构的实施,几个问题请教下: - 分割头上最后为什么还接了一层upsample,按道理conv出来的特征就是(b,n_class,h,w)已经恢复到原图大小了,那这层上采样貌似不是必要的。 - dice_loss的实现:发现dice_coef的公式用了dice_coef=inter/(pred_area + label_area - inter),但这不是iou的公式吗?dice_coef不是应该等于2*inter/(pred_area + label_area)吗?

问题类型:模型部署 **问题描述** ======================== 使用paddleX提供的deploy/cpp代码,根据官方说明修改cmakelist,修改model_infer.cpp,生成dll文件进行部署。 在c#项目加载dll后进行推理报错,跟踪dll的代码后:在如下部分报错0x00007FFDE64B742C (paddle_inference.dll)处(位于 cs_paddle_infer_dll.exe 中)引发的异常: 0xC0000005: 写入位置 0x000000000000000F 时发生访问冲突。 ``` bool PaddleInferenceEngine::Init(const InferenceConfig& infer_config) { const PaddleEngineConfig& engine_config = *(infer_config.paddle_config); paddle_infer::Config config; if ("" == engine_config.key) {...

请教下为什么线程池仅设置线程个数为1或者2?难道不应该是cpu核数吗? `ThreadPool::ThreadPool() : _active_wait(10000) { _thread_num = get_cpu_num() > 1 ? 2 : 1; }`

hello, i am learning the code of gpu memory pool and felt confusing for below code: https://github.com/OpenPPL/ppl.cv/blob/6f65a8d9e56aca1a58228e314472112363760392/src/ppl/cv/cuda/utility/memory_pool.cpp#L105 my understanding is: this while loop is to make previous iterator point to...