Open lpl071528 opened 3 months ago
Maybe the epochs are too small?
Similarly, when I train the replicated model myself and test it on the VISA dataset, the results are not as good as those of the author, with a difference of about 1 to 2 percentage points
Hello author, I tested the Visa dataset using a model that has been trained in engineering and the results were good. However, when I used the Visa dataset, created a meta.json file, and trained for 15 epochs, the testing results were very poor. What is the reason for this?
![2](https://github.com/zqhang/AnomalyCLIP/assets/85934200/1c73bb82-9f2d-481e-8c9c-05e65d3bf529)