Custom Dataset model testing with mobilenet_v2 issue

[payal211]

Hi,

I have trained mobilenet_V2 on custom dataset. when I am testing the model result is as below:

Loads checkpoint by local backend from path: ./mobilenet_v2_custom_05_05_2023/epoch_190.pth
Inference ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
{
"pred_label": 0,
"pred_score": 0.9221256971359253,
"pred_class": "tench, Tinca tinca"
}

I trained model on 6 classes. so pred_label and pred_score is good but the Pred_label is not taking the correct classes as per my classes. and it is showing the pre-defined ImageNet classes.

Can you please help me to load correct pred_classes as per actual classes.

Thanks

[Ezra-Yu]

you can try:

>>> from mmpretrain import ImageClassificationInferencer
>>> inferencer = ImageClassificationInferencer(
            model='configs/your_mobilenet_v2.py',
            weights='YOUR_CHECKPOINT_PATH.pth',
            device='cuda',
            classes=['A', 'B', C'.....])
>>> inferencer(['demo/dog.jpg', 'demo/bird.JPEG'], show_dir="./visualize/")

[payal211]

@Ezra-Yu Thanks for the quick response. 👍