thank you, please let me know how you go.
I’ve downloaded the afw.zip images, turned them into a mp4 and am getting similar output
score: [3] score: 0.207153
score: [4] score: 0.0524292
score: [5] score: 0.0820923
score: [6] score: 0.0934448
score: [7] score: 0.182495
score: [8] score: 0.227783
score: [9] score: 0.26001
score: [10] score: 0.185791
score: [11] score: 0.10199
score: [12] score: 0.09375
score: [13] score: 0.0924072
score: [14] score: 0.122253
score: [15] score: 0.124023
score: [16] score: 0.208984
score: [17] score: 0.224609
score: [18] score: 0.144897
score: [19] score: 0.120667
score: [20] score: 0.232544
score: [21] score: 0.209473
with the highest sequence looking like:
score: [69] score: 0.361816
score: [70] score: 0.610352
score: [71] score: 0.402588
score: [72] score: 0.517578
score: [73] score: 0.555176
score: [74] score: 0.577637
score: [75] score: 0.672852
but 90% of the time its like :
re: [9] score: 0.446289
score: [10] score: 0.30127
score: [11] score: 0.0866699
score: [12] score: 0.0640869
score: [13] score: 0.119507
score: [14] score: 0.103333
score: [15] score: 0.112976
score: [16] score: 0.155151
score: [17] score: 0.467529
score: [18] score: 0.270996
score: [19] score: 0.159546
score: [20] score: 0.156494
score: [21] score: 0.216187
is this also what you see ?
or is there something wrong with my hardware ?
I did noticed a few warning message in the output:
ERROR: [TRT]: 3: Cannot find binding of given name: softargmax,softargmax:1,conv_keypoints_m80
0:00:05.725646257 16898 0x5594a81930 WARN nvinfer gstnvinfer.cpp:635:gst_nvinfer_logger:<second-infer-engine1> NvDsInferContext[UID 2]: Warning from NvDsInferContextImpl::checkBackendParams() <nvdsinfer_context_impl.cpp:1868> [UID = 2]: Could not find output layer 'softargmax,softargmax:1,conv_keypoints_m80' in engine
....
0:00:06.306262305 16898 0x5594a81930 WARN nvinfer gstnvinfer.cpp:635:gst_nvinfer_logger:<primary-infer-engine1> NvDsInferContext[UID 1]: Warning from NvDsInferContextImpl::initialize() <nvdsinfer_context_impl.cpp:1161> [UID = 1]: Warning, OpenCV has been deprecated. Using NMS for clustering instead of cv::groupRectangles with topK = 20 and NMS Threshold = 0.5
do you see the same ?
the reason i ask is because i would like to know if i can depend on the facial landmarks. We do some calculation afterwards and if the source points are ‘uncertain’ to a particular degree, we may halt the calculations for a bit until the points are known to be good. but it doesn’t look like i can use the score if they are low even when the points look like they are quite accurately following the eye/nose/mouth.
even if we re-train the model to get a better model, we would still need to be able to depend on the accuracy output.
Also:
what is the output of the conv_keypoints_m80 layer ?
it says in the model card that it is 80x80x80 … but i am having some difficulty to understand what that means ?
is that some sort of confidence per pixel in the image per landmark ? what does it indicate ? is it a float ?
