mlcommons / mobile_open

MLPerf Mobile benchmarks
https://mlcommons.org/en/groups/inference-mobile/
Apache License 2.0
9 stars 1 forks source link

the results of DeepLabV3+ tflite(float and int8) predicting are wrong. #37

Open liuxpro opened 1 month ago

liuxpro commented 1 month ago

the path of used model: (https://github.com/mlcommons/mobile_models/blob/main/v0_7/tflite/deeplabv3_mnv2_ade20k_float.tflite)

the path of used predicting script : https://github.com/mlcommons/mobile_open/tree/main/vision/deeplab/models_and_code/segmentation_inference_eval_seg_tflite.py

but the result of predicting are wrong. the results of labels: array([ 2, 5, 7, 12, 13, 14, 18, 21], dtype=uint8) image

liuxpro commented 1 month ago

Who knows this reason? :D