A evaluation system for image classification
python3 run.py
Default model is gpt-4-turbo and random select one image.
Use python3 run.py -h
to check detail. Use -m
to select model to be evaluate, -n
to specify the number of images to be selected. For example, python3 run.py -m gpt-4o-mini -n 10
.
note: The dataset mini-imagenet
will be automatically downloaded when first time using it.
When running python3 run.py -m gpt-xxx
for the first time, the system will prompt you to enter your OpenAI API key. This key will be stored temporarily and is required only once. However, please note that every time you push to GitHub, a shell script will automatically remove the OpenAI API key from config/secrets.py
to ensure that sensitive information is not uploaded to the repository.
Use python3 run.py -r [int]
to set the distortion function.The distortion functions available right now are listed below:
Number | Distortion Function | available |
---|---|---|
1 | gaussian_noise | 1 |
2 | brightness | 1 |
3 | impulse_noise | 1 |
4 | speckle_noise | 1 |
5 | gaussian_blur | 1 |
6 | glass_blur | 1 |
7 | defocus_blur | 1 |
8 | zoom_blur | 1 |
9 | fog | 1 |
10 | frost | 0 |
11 | spatter | 1 |
12 | contrast | 1 |
13 | shot_noise | 1 |
14 | saturate | 1 |
15 | jpeg_compression | 0 |
16 | pixelate | 1 |
17 | elastic_transform | 1 |