search

espressif / esp-dl

Espressif deep-learning library for AIoT applications
MIT License
549 stars 118 forks source link

Evalutor only supports esp32s3? (AIV-671) #149

Closed Oreobird closed 9 months ago

Oreobird commented 9 months ago

Using latest code from master branch and quantizing model for esp32, got the error below:

Generating the quantization table: Converting coefficient to int8 per-tensor quantization for esp32 Exporting finish, the output files are: /content/drive/MyDrive/hand_recognition//trained_model//coefficient.cpp, /content/drive/MyDrive/hand_recognition//trained_model//coefficient.hpp

Quantized model info: model input name: conv2d_input, exponent: -7 Reshape layer name: StatefulPartitionedCall/sequential/conv2d/BiasAdd6, output_exponent: -7 Conv layer name: StatefulPartitionedCall/sequential/conv2d/BiasAdd, output_exponent: -8 MaxPool layer name: StatefulPartitionedCall/sequential/max_pooling2d/MaxPool, output_exponent: -8 Conv layer name: StatefulPartitionedCall/sequential/conv2d_1/BiasAdd, output_exponent: -8 MaxPool layer name: StatefulPartitionedCall/sequential/max_pooling2d_1/MaxPool, output_exponent: -8 Conv layer name: StatefulPartitionedCall/sequential/conv2d_2/BiasAdd, output_exponent: -8 MaxPool layer name: StatefulPartitionedCall/sequential/max_pooling2d_2/MaxPool, output_exponent: -8 Transpose layer name: StatefulPartitionedCall/sequential/max_pooling2d_2/MaxPool28, output_exponent: -8 Reshape layer name: StatefulPartitionedCall/sequential/flatten/Reshape, output_exponent: -8 Gemm layer name: fused_gemm_0, output_exponent: -9 Gemm layer name: fused_gemm_1, output_exponent: -9 Softmax layer name: StatefulPartitionedCall/sequential/dense_1/Softmax, output_exponent: -9

Evaluating the performance on esp32: Traceback (most recent call last): File "//content/drive/MyDrive/hand_recognition//esp_dl_quant.py", line 153, in 'minmax' if args.calib_method is None else args.calib_method) File "//content/drive/MyDrive/hand_recognition//esp_dl_quant.py", line 61, in quantize eva = evaluator.Evaluator(bw, granularity, target) File "evaluator.py", line 32, in evaluator.Evaluator.init ValueError: currently only supports esp32s3, while received esp32

Evalutor only supports esp32s3?

BlueSkyB commented 9 months ago

Yes, it's only supports esp32s3