Zero Detections using Quantization [BUG]

[dexception]

Model Name: Yolov3-Tiny Trained Resolution: 832x832 Detections using Yolo2_light: 10 Detections using Yolo2_light -quantized: 0 Detections using TensorRT: 10 Detections using TensorRT int8: 7

Output:

 Quantinization! 

 old_weight_mult = 2.000000, weights_multiplier_single = 2.000000 


 Warning: input_calibration= in the cfg-file has less values 0 than convolutional layers 0 
 Multiplers: weights 2, input 40, output 16 
 Skip layer: 3 
 old_weight_mult = 128.000000, weights_multiplier_single = 128.000000 


 Warning: input_calibration= in the cfg-file has less values 0 than convolutional layers 1 
 Multiplers: weights 128, input 40, output 0.25 
 Skip layer: 3 
 old_weight_mult = 256.000000, weights_multiplier_single = 256.000000 


 Warning: input_calibration= in the cfg-file has less values 0 than convolutional layers 2 
 Multiplers: weights 256, input 40, output 0.125 
 Skip layer: 3 
 old_weight_mult = 256.000000, weights_multiplier_single = 256.000000 


 Warning: input_calibration= in the cfg-file has less values 0 than convolutional layers 3 
 Multiplers: weights 256, input 40, output 0.125 
 Skip layer: 3 
 old_weight_mult = 256.000000, weights_multiplier_single = 256.000000 


 Warning: input_calibration= in the cfg-file has less values 0 than convolutional layers 4 
 Multiplers: weights 256, input 40, output 0.125 
 Skip layer: 3 
 old_weight_mult = 1024.000000, weights_multiplier_single = 1024.000000 


 Warning: input_calibration= in the cfg-file has less values 0 than convolutional layers 5 
 Multiplers: weights 1024, input 40, output 0.03125 
 Skip layer: 3 
 old_weight_mult = 256.000000, weights_multiplier_single = 256.000000 


 Warning: input_calibration= in the cfg-file has less values 0 than convolutional layers 6 
 Multiplers: weights 256, input 40, output 0.125 
 old_weight_mult = 1024.000000, weights_multiplier_single = 1024.000000 


 Warning: input_calibration= in the cfg-file has less values 0 than convolutional layers 7 
 Multiplers: weights 1024, input 40, output 0.03125 
 old_weight_mult = 512.000000, weights_multiplier_single = 512.000000 


 Warning: input_calibration= in the cfg-file has less values 0 than convolutional layers 8 
 Multiplers: weights 512, input 40, output 0.0625 
 old_weight_mult = 512.000000, weights_multiplier_single = 512.000000 


 Warning: input_calibration= in the cfg-file has less values 0 than convolutional layers 9 
 Multiplers: weights 512, input 40, output 0.0625 
 Skip layer: 22 
 Skip layer: 8 
 old_weight_mult = 64.000000, weights_multiplier_single = 64.000000 


 Warning: input_calibration= in the cfg-file has less values 0 than convolutional layers 10 
 Multiplers: weights 64, input 40, output 0.5 
 Skip layer: 23 
 Skip layer: 8 
 old_weight_mult = 1024.000000, weights_multiplier_single = 1024.000000 


 Warning: input_calibration= in the cfg-file has less values 0 than convolutional layers 11 
 Multiplers: weights 1024, input 40, output 0.03125 
 old_weight_mult = 128.000000, weights_multiplier_single = 128.000000 


 Warning: input_calibration= in the cfg-file has less values 0 than convolutional layers 12 
 Multiplers: weights 128, input 40, output 0.25 
 Skip layer: 22 

[ambr89]

Hi @dexception ! Did you have solved this problem?

Thanks in advance!

[jamessmith90]

You need create a calibration table.