keras-FP16-test icon indicating copy to clipboard operation
keras-FP16-test copied to clipboard

Evaluating deep learning models with float16 dtype in Keras, float16 inference

Results 5 keras-FP16-test issues
Sort by recently updated
recently updated
newest added

我用另一个VNet 使用300张dicom迭代一次就出来了大概的模型,用你这个UNET跑完之后预测图片结果全是0.5,感觉权重都没保存,但是控制台显示的准确率已经达到0.88了,按道理预测结果不会出现全部都是0.5的情况,不知道您有没有遇到这个情况

TypeError: Value passed to parameter 'scale' has DataType float16 not in list of allowed values: float32 command: eval_image_classification.py --model='densenet' --dtype='float16'

Did you do anything special to get this to work with batch normalization in Keras? Whenever I try it I get an error that float16 is not supported. Seems like...

你好,我想请问一下,fp16的模型是从哪里获取的? 是网上已有的fp16模型?还是你的工程里面有fp32转fp16的代码? 感谢!

I try to use "eval_image_classification.py" but I am not sure the meaning of "./data/val224_compressed.pkl" can you upload it or tell me where to download it. Thank you