Martin Han
Martin Han
Can you provide the code for running the model? Since you have custom objects, I am not able to run the code throguth keras2ncnn debugger :(
hmmm... I can get it running by inserting _relu6 and _hard_swish, here is the output, it seems like we get some issue with the activation :p ``` ================================== Layer Name:...
I find a bug in emitting the param file. After fix that, they are still not the same, I think the reshape is causing the issue. Still working on :p
额，应该明天能修好吧，这两天稍微有点忙
Got too busy those days. It turns out to be an easy fix. 038e61c I was able to get the correct result for a random input, can you check with...
I see you are still using the “dense_4_Softmax_blob” as the output blob, which is a bug fixed in the newer version (The version I mentioned before).  I try to...
Fixed in 34d81db, you can try the newest version.
Something I just cannot understand is that, if I injected random data, I got the same result, but if I extract the input of the ncnn, and put it into...
``` input_1 ================================== Layer Name: conv1_pad, Layer Shape: keras->(1, 225, 225, 3) ncnn->(3, 225, 225) Max: keras->1.000 ncnn->1.000 Min: keras->0.000 ncnn->0.000 Mean: keras->0.496 ncnn->0.496 Var: keras->0.291 ncnn->0.291 Cosine Similarity: 0.00000...