Open tamwaiban opened 2 years ago
thanks for your kindly reply but i have a qustion about the precision of model as you refer to. do you mean the high-precision model was trained with u2net, while the low-precision was trained with u2netp? or just the floating-point precision, the high-precision model is float32, while the low-precision is float16?
thanks for your kindly reply but i have a qustion about the precision of model as you refer to. do you mean the high-precision model was trained with u2net, while the low-precision was trained with u2netp? or just the floating-point precision, the high-precision model is float32, while the low-precision is float16?
No. the high-precision model is trained by company private data which has high-precision lable. the low-precision was trained by ADE20K datasets. the size of float16 model is half of float32 model and the precision almost dosen't decrease if inferenced by ncnn.
@tamwaiban Thank you so much for the project, it was a great job! Could you please provide the download link of the data set? I am a graduate student from China. I failed to download data set in sky-Optimization project.(https://github.com/google/sky-optimization) I wonder if you can provide additional download links, such as Baidu Cloud
@tamwaiban Thank you so much for the project, it was a great job! Could you please provide the download link of the data set? I am a graduate student from China. I failed to download data set in sky-Optimization project.(https://github.com/google/sky-optimization) I wonder if you can provide additional download links, such as Baidu Cloud
I did't put the datasets on Baidu Cloud and I had lose the datasets. perhaps you had better get a VPN. you can download the orignal ADE20K datasets mask or label the datasets by my model
Thanks for your reply, you provided a good idea!
郝大为 | |
---|---|
@. | On 5/11/2022 @.> wrote:
@tamwaiban Thank you so much for the project, it was a great job! Could you please provide the download link of the data set? I am a graduate student from China. I failed to download data set in sky-Optimization project.(https://github.com/google/sky-optimization) I wonder if you can provide additional download links, such as Baidu Cloud
I did't put the datasets on Baidu Cloud and I had lose the datasets. perhaps you had better get a VPN. you can download the orignal ADE20K datasets mask or label the datasets by my model
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>
@tamwaiban the ncnn model couln't be convert to pytorch model. I can upload a pytorch model that the precision is not very good for test right now.
if you want to get the result as the repo I suggest you train the model by yourself. the heigh-precision model is used in our commercial products, I couldn't upload it