TensorRT 2.0 EA sample int8 error?

I run sample_int8 ,it shows:

INT8 run:400 batches of size 100 starting at 100 with cutoff 1 and quantile index 6
sample_int8: sampleINT8.cpp:58: std::string locateDataDir(): Assertion `i != MAX_DEPTH’ failed

my data directory don’t have int8 data file.

how to get the int8 data file

I only have this file:
./data:
total 12
drwxr-xr-x 3 root root 4096 5月 5 14:55 ./
drwxr-xr-x 10 root root 4096 5月 5 14:56 …/
drwxr-xr-x 4 root root 4096 5月 5 14:55 samples/

./data/samples:
total 16
drwxr-xr-x 4 root root 4096 5月 5 14:55 ./
drwxr-xr-x 3 root root 4096 5月 5 14:55 …/
drwxr-xr-x 2 root root 4096 5月 5 14:55 googlenet/
drwxr-xr-x 2 root root 4096 5月 5 14:55 mnist/

./data/samples/googlenet:
total 40556
drwxr-xr-x 2 root root 4096 5月 5 14:55 ./
drwxr-xr-x 4 root root 4096 5月 5 14:55 …/
-rw-r–r-- 1 root root 41481639 10月 18 2016 googlenet.caffemodel
-rw-r–r-- 1 root root 36261 10月 18 2016 googlenet.prototxt

./data/samples/mnist:
total 16760
drwxr-xr-x 2 root root 4096 5月 5 14:55 ./
drwxr-xr-x 4 root root 4096 5月 5 14:55 …/
-rw-r–r-- 1 root root 797 10月 18 2016 0.pgm
-rw-r–r-- 1 root root 797 10月 18 2016 1.pgm
-rw-r–r-- 1 root root 797 10月 18 2016 2.pgm
-rw-r–r-- 1 root root 797 10月 18 2016 3.pgm
-rw-r–r-- 1 root root 797 10月 18 2016 4.pgm
-rw-r–r-- 1 root root 797 10月 18 2016 5.pgm
-rw-r–r-- 1 root root 797 10月 18 2016 6.pgm
-rw-r–r-- 1 root root 797 10月 18 2016 7.pgm
-rw-r–r-- 1 root root 797 10月 18 2016 8.pgm
-rw-r–r-- 1 root root 797 10月 18 2016 9.pgm
-rw-r–r-- 1 root root 7685475 10月 18 2016 mnistapi.wts
-rw-r–r-- 1 root root 1725135 10月 18 2016 mnist.caffemodel
-rw-r–r-- 1 root root 7685475 10月 18 2016 mnistgie.wts
-rw-r–r-- 1 root root 3147 10月 18 2016 mnist_mean.binaryproto
-rw-r–r-- 1 root root 1806 10月 18 2016 mnist.prototxt

I met the same problem with you. Have you solved it? I use images as input data to calibrate my caffemodel, it can be worked, but the result is wrong. I don’t know why.

@Mickle_chen
sorry I don’t know how to solve it.Wait for Nvidia.

maybe you can try to use absolute path but not “…/data/int8” in the source code

and the the batches for using INT8 in the sample_int8 are in $TensorRT_Root/data/mnist/batches