Weiqian Chen
Weiqian Chen
The first layer is used to convert the images to spike. To be specific, we use the IF neurons with the images as the input to generate spike.
font{ line-height: 1.6; } ul,ol{ padding-left: 20px; list-style-position: inside; } 目前可以做到的极限是，多块gpu每块将batch_size设为1，这样可以做大程度压缩mini-batch，然后在此基础上增大timesteps至每块gpu的显存上限。 cwq ***@***.*** 签名由 网易邮箱大师 定制 On 06/3/2021 ***@***.***> wrote： 作者您好！ 我单块GPU的显存上限只有12g，在运行ann-to-snn的时候，只能设置的time step值为16，否则就会显存不够，然而这样的值转换出来的SNN的mAP聊胜于无。 我能否通过使用多块gpu的方式来设置更高的time steps参数呢？如果有，请您告诉我。 非常感谢您的帮助。 —You are receiving this...
train on VOC07+12 test on VOC2007 no need to be changed about the training config
font{ line-height: 1.6; } ul,ol{ padding-left: 20px; list-style-position: inside; } Have you tried other datasets? Please show your training process in detail. Weiqian Chen [email protected] 签名由 网易邮箱大师 定制 On 01/10/2021...
For training problems, you can refer to ultralytics/yolov3. Maybe you should revise datasets.py in /utils or the code for data preprocessing(such as [0,255] to [0,1]).
almost two months ago
font{ line-height: 1.6; } ul,ol{ padding-left: 20px; list-style-position: inside; } Please use PyTorch1.3 Weiqian Chen [email protected] 签名由 网易邮箱大师 定制 On 01/14/2021 22:53，tomorrow-mouse wrote： I have found something weird. I firstly...
Please use PyTorch1.3
If the input is a SpikeTensor, which is a time series of 0 or 1, then it should be maintained.
font{ line-height: 1.6; } ul,ol{ padding-left: 20px; list-style-position: inside; } Sorry, I haven’t coded to load the transformed snn directly for inference, but just to test the mAP. I will...