leafjungle

我理解是：这里不仅做了压缩，还做了格式转换（从png到jpg）。

I print the time cost for each step, and the result: image size: 100K. time cost: total cost: 86 seconds. ApplyGlobalQuantization: 20 seconds. SelectFrequencyMasking: 60 seconds.

where is cu_mem defined? cuda or opencl can not find that.

@pornel , is there an experiment shows the Butteraugli scores of guetzli and libjpeg? (based on the same quality)

Currently I think CUDA is popular enough. And with specified version of CUDA driver, I think it is useful to do that.

@pornel , yeah, that sounds great. I git clone the from http://github.com/ianhuang-777/guetzli. And I can not found cu_mem defined in anywhere of cuda or opencl.

It may be caused by: the compile version and spark run version is different(class : org/apache/spark/ml/param/ParamMap). But I have checked them. compile version: 2.11.12 2.11 spark version is:  So,...

@searchivarius ， Great. is there any example to add new data to exist index by c++?

@searchivarius , that means HNSW support deletion and addition one by one now? that is quite useful for online service to add/del data immediately.

@Tabrizian , thank you for your kindly reply. yes, python2.7 is old. but python with version < 3.8 also can not be deployed in triton as a python backend. Many...