Description
I am trying to optimise my Mask-RCNN model for better and faster inference. For this, I converted my .h5 model`to UFF using the instructions given in the TensorRT repo. The environment I used was:
CUDA: 10.0
cudnn: 7.6.2
TensorRT: 7.0.0.11
I successfully converted both the sample model and my custom model into UFF.
Then, I built the samples in another environment. This was because the first environment caused some errors. This time I used:
CUDA: 10.2
cudnn: 8.1
TensorRT: 7.0.0.11
I was able to compile and build the sample and even run it. For my custom model (now UFF), I made relevant changes to the mrcnn_config.h
diff --git a/content/TensorRT/samples/opensource/sampleUffMaskRCNN/backup_mrcnn_config.h b/content/TensorRT/samples/opensource/sampleUffMaskRCNN/mrcnn_config.h
index cf24cbc..746c294 100644
--- a/content/TensorRT/samples/opensource/sampleUffMaskRCNN/backup_mrcnn_config.h
+++ b/content/TensorRT/samples/opensource/sampleUffMaskRCNN/mrcnn_config.h
@@ -13,7 +13,6 @@
* See the License for the specific language governing permissions and
* limitations under the License.
*/
-
#ifndef MASKRCNN_CONFIG_HEADER
#define MASKRCNN_CONFIG_HEADER
#include "NvInfer.h"
@@ -57,7 +56,7 @@ static const int FPN_CLASSIF_FC_LAYERS_SIZE = 1024;
static const int TOP_DOWN_PYRAMID_SIZE = 256;
// Number of classification classes (including background)
-static const int NUM_CLASSES = 1 + 80; // COCO has 80 classes
+static const int NUM_CLASSES = 1 + 3; // COCO has 80 classes
// Length of square anchor side in pixels
static const std::vector<float> RPN_ANCHOR_SCALES = {32, 64, 128, 256, 512};
@@ -85,86 +84,9 @@ static const int POST_NMS_ROIS_INFERENCE = 1000;
// COCO Class names
static const std::vector<std::string> CLASS_NAMES = {
"BG",
- "person",
- "bicycle",
- "car",
- "motorcycle",
- "airplane",
- "bus",
- "train",
- "truck",
- "boat",
- "traffic light",
- "fire hydrant",
- "stop sign",
- "parking meter",
- "bench",
- "bird",
- "cat",
- "dog",
- "horse",
- "sheep",
- "cow",
- "elephant",
- "bear",
- "zebra",
- "giraffe",
- "backpack",
- "umbrella",
- "handbag",
- "tie",
- "suitcase",
- "frisbee",
- "skis",
- "snowboard",
- "sports ball",
- "kite",
- "baseball bat",
- "baseball glove",
- "skateboard",
- "surfboard",
- "tennis racket",
- "bottle",
- "wine glass",
- "cup",
- "fork",
- "knife",
- "spoon",
- "bowl",
- "banana",
- "apple",
- "sandwich",
- "orange",
- "broccoli",
- "carrot",
- "hot dog",
- "pizza",
- "donut",
- "cake",
- "chair",
- "couch",
- "potted plant",
- "bed",
- "dining table",
- "toilet",
- "tv",
- "laptop",
- "mouse",
- "remote",
- "keyboard",
- "cell phone",
- "microwave",
- "oven",
- "toaster",
- "sink",
- "refrigerator",
- "book",
- "clock",
- "vase",
- "scissors",
- "teddy bear",
- "hair drier",
- "toothbrush",
+ "panel",
+ "back",
+ "dock"
};
static const std::string MODEL_NAME = "mrcnn_nchw.uff";
@@ -174,4 +96,4 @@ static const std::vector<std::string> MODEL_OUTPUTS = {"mrcnn_detection", "mrcnn
static const Dims2 MODEL_DETECTION_SHAPE{DETECTION_MAX_INSTANCES, 6};
static const Dims4 MODEL_MASK_SHAPE{DETECTION_MAX_INSTANCES, NUM_CLASSES, 28, 28};
} // namespace MaskRCNNConfig
-#endif
\ No newline at end of file
+#endif
I recompile everything and it goes well. But now, when I try to run the executable on my custom model, I get the following error:
&&&& RUNNING TensorRT.sample_maskrcnn # ./sample_uff_maskRCNN --datadir /content/TensorRT-7.0.0.11/data/faster-rcnn --fp16
[05/24/2021-10:58:34] [I] Building and running a GPU inference engine for Mask RCNN
[05/24/2021-10:58:36] [I] [TRT] UFFParser: Did not find plugin field entry image_size in the Registered Creator for layer ROI
sample_uff_maskRCNN: detectionLayerPlugin.cpp:221: void nvinfer1::plugin::DetectionLayer::check_valid_inputs(const nvinfer1::Dims*, int): Assertion `inputs[0].nbDims == 4 && inputs[0].d[1] == mNbClasses * 4' failed.
---------------------------------------------------------------------------
CalledProcessError Traceback (most recent call last)
<ipython-input-60-6d09927647ae> in <module>()
1
----> 2 get_ipython().run_cell_magic('shell', '', './sample_uff_maskRCNN --datadir $TRT_DATADIR --fp16')
2 frames
/usr/local/lib/python3.7/dist-packages/google/colab/_system_commands.py in check_returncode(self)
137 if self.returncode:
138 raise subprocess.CalledProcessError(
--> 139 returncode=self.returncode, cmd=self.args, output=self.output)
140
141 def _repr_pretty_(self, p, cycle): # pylint:disable=unused-argument
CalledProcessError: Command './sample_uff_maskRCNN --datadir $TRT_DATADIR --fp16' died with <Signals.SIGABRT: 6>.
Environment
Google Colaboratory
TensorRT Version: 7.0.0.11:
GPU Type: Tesla P100-PCIE-16GB:
Nvidia Driver Version:
CUDA Version: specified above:
CUDNN Version: specified above:
Operating System + Version: Ubuntu 18.04.5 LTS:
Python Version (if applicable): 3.7:
TensorFlow Version (if applicable): 1.5-gpu
Relevant Files
Colab/ipynb file: (removed)
Custom UFF file: link
mrcnn_config.h: mrcnn_config.h (3.6 KB)
Steps To Reproduce
The steps have been mentioned in the iPython notebook attached above.
Any help is appreciated.