03d2b88fa6
Add TKDNN_CALIB_IMG_PATH and TKDNN_CALIB_LABRL_PATH variable
...
This commit adds two variables for the calibration dataset.
The first is reffered to .txt file that contains the list
of the absolute paths of the images for the INT8 calitration.
The second is referred to .txt file that contains the list
of the absolute paths of the labels of the same images above.
Signed-off-by: Davide Sapienza <sapienza.dav@gmail.com >
2020-03-30 18:55:32 +02:00
0474f019cf
Add TKDNN_MODE variable to the name of the network.
...
This commit permits to obtain different .rt files for different
precision optimizations of the same network.
Signed-off-by: Davide Sapienza <sapienza.dav@gmail.com >
2020-03-30 15:08:25 +02:00
e540213da6
Add the Int8 calibrator and the tensorRT Int8 inference
...
Signed-off-by: Davide Sapienza <sapienza.dav@gmail.com >
2020-03-27 00:48:14 +01:00
785a3d8dd5
Add GPU version for yolo3detection
...
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com >
2020-03-23 21:29:29 +01:00
3bc648c07b
Modify map demo, using abstract class. Move draw function in abstract class
...
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com >
2020-03-23 19:32:37 +01:00
8b9516da97
Refactoring for detection NN
...
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com >
2020-03-20 21:14:12 +01:00
997e857e32
Add OPENCV_CUDA define, to allow having preprocess both in CPU and GPU (Mobilenet and Centernet)
...
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com >
2020-03-20 10:21:11 +01:00
5c6d6ae0af
Style fix, useless params removed
...
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com >
2020-03-19 18:13:35 +01:00
cae0b84687
Add preprocess function, allow preprocess on GPU for mobilenetdetection
...
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com >
2020-03-19 17:53:29 +01:00
1a50065bde
Fix conv2d with additional bias for tensorRT. Fix reshape deserialize. Mb2512 works with tensorRT
...
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com >
2020-03-17 13:18:05 +01:00
14d850751d
Add mobilenetv2ssdlite512 test.
...
Works for CuDNN, not for tensorRT.
Modified channels in second convolution for classification headers from 126 to 486, when changing size from 300 to 512.
Added size 512 SSD specs and support to COCO dataset (81 classes, first BACKGROUND due to repo for training).
Refactoring class names.
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com >
2020-03-17 10:35:02 +01:00
677b6b2698
Refactoring
...
Signed-off-by: Francesco Gatti <gattifrancesco@hotmail.it >
2020-03-14 17:22:58 +01:00
21b8146232
Add function to download weigths if do no exist, for each test. Add some controls in map demo.
...
Signed-off-by: Francesco Gatti <gattifrancesco@hotmail.it >
2020-03-14 17:02:07 +01:00
a65c8aa91e
Merge branch 'cnet' of https://github.com/ceccocats/tkDNN into cnet
2020-03-12 20:18:11 +01:00
4845edbc88
csresnext50-panet-spp works with TensorRT
...
Signed-off-by: Francesco Gatti <gattifrancesco@hotmail.it >
2020-03-12 18:24:41 +01:00
5f4508a5a6
Add csresnext50-panet-spp test. Works on CUDNN. Does not work with TensorRT
...
Signed-off-by: Francesco Gatti <gattifrancesco@hotmail.it >
2020-03-11 18:37:26 +01:00
98770ee0c5
Merge
...
Signed-off-by: xavier <micaelaverucchi@gmail.com >
2020-03-10 15:43:26 +01:00
07a61b0e37
Add tiny yolo 512 test
...
Signed-off-by: xavier <micaelaverucchi@gmail.com >
2020-03-10 15:40:11 +01:00
75d5e20843
Add writing results on file for map demo
...
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com >
2020-03-10 14:35:25 +01:00
6fee8d5ec5
- ImuOdom model into class
...
- Route layer input array hard copy
- fix utils
2020-03-09 19:44:21 +01:00
c94159a2bf
Add draw method to Yolo3Detection class.
...
Update yolo demo
Signed-off-by: Davide Sapienza <sapienza.dav@gmail.com >
2020-03-03 17:08:25 +01:00
1d38d88c85
Fix dependencies problems
...
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com >
2020-02-27 16:48:10 +01:00
3be254b08f
Add Mobilenetv2 SSD Lite post and preprocessing, add mobilenet demo
...
Signed-off-by: xavier <micaelaverucchi@gmail.com >
2020-02-26 17:42:20 +01:00
c4cef1aa02
Merge branch 'cnet' of https://github.com/ceccocats/tkDNN into cnet
2020-02-21 10:49:35 +01:00
808a84131c
Add Mobilenet2SSDLite test
...
The new test works both with TensorRT and cuDNN. Preprocessing and
Postprocessing are missing. Add ClippedReLU (for ReLU6), groups for
Conv2d, additional bias for convolution.
Other minors:
-move the timer in the detector to measure all the
processing time for a given frame (both centernet and yolo);
-add int8 flag.
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com >
Davide Sapienza <sapienza.dav@gmail.com >
2020-02-21 10:45:46 +01:00
cd1e1c1582
LSTM ok
2020-02-16 17:21:58 +01:00
4b48a2f38c
structure ok, result wrong
2020-02-16 17:08:19 +01:00
0ac292ea48
works but it need cleaning
2020-02-16 16:28:39 +01:00
8736a8c3da
LSTM params
2020-02-15 20:37:08 +01:00
200e9466f6
Move pre-processing on GPU
...
Signed-off-by: Davide Sapienza <sapienza.dav@gmail.com >
2020-02-14 18:44:13 +01:00
f5706d8523
lstm return seq
2020-02-13 23:21:28 +01:00
0329318ca3
removed unused var
2020-02-13 23:10:48 +01:00
e8280d3a65
LSTM to be tested
2020-02-13 23:04:29 +01:00
c876fa05ae
LSTM cudnn test
2020-02-13 19:27:18 +01:00
40a4a55cd0
Fix a bug in the deformable kernel.
...
There was a wrong variable initialization.
Fixes: 51ffcb1f50sapienza.dav@gmail.com >
2020-02-11 16:28:17 +01:00
d0026ff235
Change README.md, add config.yaml, refactoring
...
Signed-off-by: xavier <micaelaverucchi@gmail.com >
2020-02-11 11:25:42 +01:00
e01d7c6b13
Merge branch 'cnet' of https://github.com/ceccocats/tkDNN into cnet
2020-02-11 10:32:52 +01:00
2b06836a7f
Optimize deformable kernel
...
Signed-off.by: Ignacio Sañudo Olmedo<ignacio.sanudoolmedo@unimore.it >
Signed-off-by: Davide Sapienza <sapienza.dav@gmail.com >
2020-02-11 10:06:08 +01:00
c7cb3e7571
Add avg precision, recall, f1score computation, other minors
...
Signed-off-by: xavier <micaelaverucchi@gmail.com >
2020-02-10 18:19:08 +01:00
17a083ae28
Remove mallocs and frees from the kernels
...
Signed-off-by: Davide Sapienza <sapienza.dav@gmail.com >
2020-02-07 14:50:05 +01:00
da24c2e752
Refactoring map_demo, add evaluation.h and evaluation.cpp
...
Signed-off-by: xavier <micaelaverucchi@gmail.com >
2020-02-07 12:47:02 +01:00
5abf1613f2
Fix memory leak
...
This commit moves cublasCreate out from dcn_v2_cuda_forward
to save some milliseconds and it adds cublasDestroy (cause
of memory leak).
Signed-off-by: Davide Sapienza <sapienza.dav@gmail.com >
2020-02-06 23:09:12 +01:00
516c1e6636
Change CenterNet input dimension.
...
This commit changes the image input dimension, it updates the
CenterNet detection class.
Signed-off-by: Davide Sapienza <sapienza.dav@gmail.com >
2020-02-06 18:08:06 +01:00
0d6acaec41
Add grouped convolutions in CUDNN and tensorRT.
...
Signed-off-by: Davide Sapienza <sapienza.dav@gmail.com >
2020-02-05 14:32:58 +01:00
07da54efe9
Centernet: fix pooling problem, add centrnet demo
...
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com >
Signed-off-by: Davide Sapienza <sapienza.dav@gmail.com >
2020-01-21 12:50:18 +01:00
b7a6e8c8f5
Merge branch 'master' of https://github.com/ceccocats/tkDNN into cnet
2020-01-20 16:41:19 +01:00
6a3b261bc9
compile with tensorrt 5
2020-01-20 14:51:38 +01:00
b23cf9fe70
Merge branch 'cnet' of https://github.com/ceccocats/tkDNN into cnet
...
Signed-off-by: Michaela Verucchi <micaelaverucchi@gmail.com >
Signed-off-by: Davide Sapienza <sapienza.dav@gmail.com >
2020-01-20 12:35:27 +01:00
7838cb4922
Pre-process, Process and Post-process work
...
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com >
Signed-off-by: Davide Sapienza <sapienza.dav@gmail.com >
2020-01-20 12:27:49 +01:00
da4f246157
add DLA, plugin for shortcut and leaky. new verison 0.4
2020-01-15 21:48:18 +01:00
Davide Sapienza
Micaela Verucchi
Francesco Gatti
luca