How to use SAHI to enhance the detection effect of my own yolov5 model

[nemolfie]

After reading our discussion, we all installed yolov5 in anaconda. But I hope to add SAHI to the original Yolov5 detect.py, or use other methods to enhance the detection effect of my model. I hope that some kind-hearted people can provide the following examples or ideas on how to join SAHI

[fcakyon]

check this yolov5+sahi notebook: https://github.com/obss/sahi/blob/main/demo/inference_for_yolov5.ipynb

[mwitiderrick]

Hello @fcakyon can I use this notebook with a pruned and quantized YOLOv5 model? https://sparsezoo.neuralmagic.com/models/cv%2Fdetection%2Fyolov5-s%2Fpytorch%2Fultralytics%2Fcoco%2Fpruned85_quant-none

I tried and got this error

WARNING ⚠️ DetectMultiBackend failed: 'collections.OrderedDict' object has no attribute 'to'
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
[/usr/local/lib/python3.9/dist-packages/yolov5/helpers.py](https://localhost:8080/#) in load_model(model_path, device, autoshape, verbose, hf_token)
     37     try:
---> 38         model = DetectMultiBackend(
     39             model_path, device=device, fuse=autoshape, hf_token=hf_token

8 frames
AttributeError: 'collections.OrderedDict' object has no attribute 'to'

During handling of the above exception, another exception occurred:

AttributeError                            Traceback (most recent call last)
AttributeError: 'collections.OrderedDict' object has no attribute 'to'

During handling of the above exception, another exception occurred:

TypeError                                 Traceback (most recent call last)
[/usr/local/lib/python3.9/dist-packages/sahi/models/yolov5.py](https://localhost:8080/#) in load_model(self)
     30             self.set_model(model)
     31         except Exception as e:
---> 32             raise TypeError("model_path is not a valid yolov5 model path: ", e)
     33 
     34     def set_model(self, model: Any):

TypeError: ('model_path is not a valid yolov5 model path: ', AttributeError("'collections.OrderedDict' object has no attribute 'to'"))

[kadirnar]

Can you share the prediction codes of this model?
https://docs.neuralmagic.com/use-cases/object-detection/deploying

Have you tried these codes? I can add this support to SAHI library.

[mwitiderrick]

@kadirnar We would like to add support for Deepsparse on SAHI. Is it possible? This is how the inference code looks like

pip install deepsparse 
wget -O basilica.jpg https://raw.githubusercontent.com/neuralmagic/deepsparse/main/src/deepsparse/yolo/sample_images/basilica.jpg
from deepsparse import Pipeline

model_stub = "zoo:cv/detection/yolov5-l/pytorch/ultralytics/coco/pruned-aggressive_98"
images = ["basilica.jpg"]

yolo_pipeline = Pipeline.create(
    task="yolo",
    model_path=model_stub,
)

pipeline_outputs = yolo_pipeline(images=images, iou_thres=0.6, conf_thres=0.001)

Essentially we load a sparse-quantized YOLOv5 model from Neural Magic's SparseZoo.

Here's a live demo
https://huggingface.co/spaces/neuralmagic/cv-yolo

@kadirnar, the model stub is obtained from https://sparsezoo.neuralmagic.com/?repo=ultralytics&page=1