Dear Sir,
I had replaced the mmalib_01_00_00_00 with mmalib_01_01_00_02.Made Changes in config.mk in tidl/makerules and psdkra_tools.mk in tiovx. Whole sdk clean build done.
1.Is the process right?.
2.If so, I had tried both configurations with numParamBits and numFeatureBits both set to 8 and 16.
In numParamBits and numFeatureBits set to 8 configuration ,observed misses and false detection in few frames which was not observed in the previous release.
a.
b)
3.With NumParamBits and NumFeatureBits set to 16,observed multiple bounding boxes and misses.
a)
Also the import console out for 16 bit is
----------------------- TIDL Process with REF_ONLY FLOW ------------------------ # 0 . .. T 3367.98 ... .... ..... ------------------ Network Compiler Traces ----------------------------- Main iteration numer: 0.... Preparing for memory allocation : internal iteration number: 0 successful Memory allocation -------------------- Network Compiler : Analysis Results are available -------------------- SUGGESTION: [TIDL_BatchNormLayer] data/bias 16 bits is not optimal in this release. SUGGESTION: [TIDL_FlattenLayer] ctx_output5/sep/relu_mbox_loc_flat Flatten with 16 bit is not optimal in this version. SUGGESTION: [TIDL_FlattenLayer] ctx_output5/sep/relu_mbox_conf_flat Flatten with 16 bit is not optimal in this version. SUGGESTION: [TIDL_FlattenLayer] ctx_output4/sep/relu_mbox_loc_flat Flatten with 16 bit is not optimal in this version. SUGGESTION: [TIDL_FlattenLayer] ctx_output4/sep/relu_mbox_conf_flat Flatten with 16 bit is not optimal in this version. SUGGESTION: [TIDL_FlattenLayer] ctx_output3/sep/relu_mbox_loc_flat Flatten with 16 bit is not optimal in this version. SUGGESTION: [TIDL_FlattenLayer] ctx_output3/sep/relu_mbox_conf_flat Flatten with 16 bit is not optimal in this version. SUGGESTION: [TIDL_FlattenLayer] ctx_output2/sep/relu_mbox_loc_flat Flatten with 16 bit is not optimal in this version. SUGGESTION: [TIDL_FlattenLayer] ctx_output2/sep/relu_mbox_conf_flat Flatten with 16 bit is not optimal in this version. SUGGESTION: [TIDL_FlattenLayer] ctx_output1/sep/relu_mbox_loc_flat Flatten with 16 bit is not optimal in this version. SUGGESTION: [TIDL_FlattenLayer] ctx_output1/sep/relu_mbox_conf_flat Flatten with 16 bit is not optimal in this version. **************************************************** ** ALL MODEL CHECK PASSED ** ****************************************************
Note:I'm using tidl version: tidl_j7_01_00_00_00.
Kindly guide us
Thanks & Regards


