YOLOv7 to TFlite conversion input tensor error

Community Help
1

I am using Flutter in Android Studio. I trained a model in YOLOv7 (Roboflow) and I converted the model to TFlite in Google Colab with this website as reference: Export Yolo V7 to Tensorflow Lite

My Colab code:

!pip install onnx onnxruntime onnxsim onnx-tf
!git clone https://github.com/WongKinYiu/yolov7.git
%cd yolov7
!wget https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7.pt
!pip install -r requirements.txt
from google.colab import drive
drive.mount('/content/gdrive')
!cp /content/gdrive/MyDrive/best.pt /content/yolov7
!python export.py --weights /content/yolov7/best.pt --grid --end2end --simplify \
        --topk-all 100 --iou-thres 0.65 --conf-thres 0.35 --img-size 640 640 --max-wh 640
!pip install tensorflow tensorflow_probability
!mkdir tfmodel
!onnx-tf convert -i best.onnx -o tfmodel/
import tensorflow as tf
converter = tf.lite.TFLiteConverter.from_saved_model('tfmodel/')
tflite_model = converter.convert()
with open('tfmodel/best.tflite', 'wb') as f:
    f.write(tflite_model)

After deploying it on my object detection app, I get the following error: Failed to initialize detector. The input tensor should have dimensions 1 x height x width x 3. Got 1 x 3 x 640 x 640., null, null).
I believe the issue is the channel? How should I fix this error?

2

Hi @RatedEdge

In general, we don’t provide support for notebooks and code that isn’t related to Roboflow or its products.

That being said, yes it does seem to be a channel issue. It might be somewhere in the conversion that this issue occurred.

If you’d like to use Roboflow, you can try our TFLite guide or try training with Roboflow and deploying to whatever platform you’d like.

4

Is there a solution for this error?

image
image2588×1042 263 KB

5

Can you try running:

%tensorflow_version 1.15
7

I ran it and got this:

image
image1035×441 56.4 KB