the tflite file which pass vela the result is different from the tflite file without passing vela

Hi,

I use the tflite file and pass the vela (3.9.0 & 3.10.0) and run on ml-embedded-evaluation-kit to see the result. Finally, the result of output tensor is really different from the tflite without passing vela.

the file `yolox_ti_body_head_hand_t_1x3x128x160_bgr_uint8.tflite` is the original tflite file.

`yolox_ti_body_head_hand_n_1x3x128x160_bgr_uint8_vela.tflite` is the tflite file which pass vela 3.9.0

`yolox_ti_body_head_hand_n_1x3x128x160_bgr_uint8_vela_3_10_0.tflite` is the tflite file which pass vela 3.10.0

I run the same image on python interpreter using `yolox_ti_body_head_hand_t_1x3x128x160_bgr_uint8.tflite` which result is the same as on ml-embedded-evaluation-kit using `yolox_ti_body_head_hand_t_1x3x128x160_bgr_uint8.tflite`.

But while using the tflite passing vela, the result is really different.

Do you have any idea?

Sincerely,

Kris

Parents Reply Children
No data