search

nvdla / sw

NVDLA SW
Other
482 stars 191 forks source link

Can I run image classfication with cifar10_quick? #44

Closed MINZHIJI closed 6 years ago

MINZHIJI commented 6 years ago

Environment:

Question: I run cifar10_quick which is compiled by nvdla_compiler and I use nvdla_runtime to run cifar10_quick. But the runtime is no response for a long time! [Solved] New problem is the output results are similar! I try three cases! Results

[test/0_cat.jpg]
result:
0.0689087 0.0415649 0.134766 0.111938 0.219238 0.0861206 0.10498 0.0811768 0.100342 0.0509338 

[train/27_dog.jpg]
result:
0.0667114 0.0420837 0.136597 0.111877 0.220703 0.086792 0.106567 0.079895 0.0982056 0.0506287 

[nope]    //   no input image
result:
0.0683594 0.0415344 0.140625 0.111206 0.224121 0.0821533 0.105042 0.0791626 0.097229 0.0505676
JunningWu commented 6 years ago

2 HWLs done, totally 16 layers

MINZHIJI commented 6 years ago

@JunningWu Do yo finish cifar10_quick? My NVDLA stops at 1 HWLs!

JunningWu commented 6 years ago

@MINZHIJI have you tried some other resolution image, 3x227x227? can you share your caffemodel?

MINZHIJI commented 6 years ago

CIFAR10 Environment (Link) @JunningWu I upload to google drive. There are input image, deploy.prototxt, caffemodel, and compiled loadable file.

JunningWu commented 6 years ago

@MINZHIJI the simulation result is 0.0689087 0.0415649 0.134766 0.111938 0.219238 0.0861206 0.10498 0.0811768 0.100342 0.0509338

and the process ends very quickly. the test image is 0_cat.jpg

update 16_dog.jpg result

0.0703735 0.0430908 0.134399 0.112061 0.213379 0.0882568 0.105408 0.0809326 0.0998535 0.0523682

MINZHIJI commented 6 years ago

@JunningWu
Which is input image? And are you use my provided loadable or yourself compiled? Finally, thanks you for your testing.

MINZHIJI commented 6 years ago

@JunningWu Thanks. I use new runtime to run cifar10 sucessfully. But output results are very strange. Result values are very similar!? I try three cases! Results

[test/0_cat.jpg]
result:
0.0689087 0.0415649 0.134766 0.111938 0.219238 0.0861206 0.10498 0.0811768 0.100342 0.0509338 

[train/27_dog.jpg]
result:
0.0667114 0.0420837 0.136597 0.111877 0.220703 0.086792 0.106567 0.079895 0.0982056 0.0506287 

[nope]    //   no input image
result:
0.0683594 0.0415344 0.140625 0.111206 0.224121 0.0821533 0.105042 0.0791626 0.097229 0.0505676
JunningWu commented 6 years ago

@MINZHIJI yeah，the simulation results are the same for 0_cat.jpg.

prasshantg commented 6 years ago

@MINZHIJI is this issue resolved with latest changes?

MINZHIJI commented 6 years ago

@prasshantg yes, thanks.