Crazod

> > Vulkan Backend 没有 auto-tuning 或者针对不同机型调 local size，计算量较小的模型确实有可能出现高端 GPU 反而更慢的情况的 那请问我该如何能跑benchmark 中的mi 6 的数据呢，比如mobilenetv2 只有10ms左右的数据。

Hi, I have use the l2_loss and CE loss: Both of them can get similar result.(90%~) However, i notice that most of the dark region (no response region)in l2_loss will...

Hi recently i found a paper. https://arxiv.org/abs/1703.00862 Binarized Convolutional Landmark Localizers for Human Pose Estimation and Face Alignment with Limited Resources. It also use cse and present some explainment.

> hi, I meet the same problem, Can you share the related code of MSE loss? Thank you @dongzhuoyao @Crazod Sorry, this project is too old to find, loss may...

Hi, i have figure out that why the transform.rotate(img, r_angle, preserve_range = True) get errors. The reason is the version of skimage(sickit image). It seems only high version sickit image...

HI @beckyng2194 I add code at inference.py `if __name__ == '__main__': params = process_config('config.cfg') dataset = DataGenerator(params['joint_list'], params['img_directory'], params['training_txt_file'], remove_joints=params['remove_joints']) infer = Inference() infer.webcamSingle() ` webcamYOLO is similiar

Hi I have follow @o0t1ng0o steps. And down load the pretained model(refined and tiny). I have I convert the test_img and key point. But it seems still a wrong result....

@GajjarMihir Hi, i have figure out that why the transform.rotate(img, r_angle, preserve_range = True) get errors. The reason is the version of skimage(sickit image). It seems only high version sickit...

> 如果把 kl 的公式展开，和 student 无关的项省掉后（teacher 部分不影响梯度更新，只影响数值），就是代码里的形式 明白了，多谢作者。

> 使用train中的代码进行蒸馏训练，看日志loss开始会下降一点，后面一致在44.几震荡，感觉不收敛，是设置的不对吗 Hi, 我使用了他的代码，出现了类似的问题。是因为代码中使用的Teacher权重不能正确加载导致。你可以修改 [mmseg/models/backbones/resnet.py](https://github.com/pppppM/mmsegmentation-distiller/blob/master/mmseg/models/backbones/resnet.py#L674) deep_stem 为False，就可以正确加载权重了。我使用单卡训练到16000/80000。 +--------+-------+-------+-------+ | Scope | mIoU | mAcc | aAcc | +--------+-------+-------+-------+ | global | 44.94 | 53.51 | 89.52 | +--------+-------+-------+-------+ Loss 为...