search

FuNz-0 / PromptAD

Official implementation of CVPR 2024 PromptAD: Learning Prompts with Only Normal Samples for Few-Shot Anomaly Detection
The Unlicense
92 stars 5 forks source link

Runtime and Results #21

Open kirnde opened 3 days ago

kirnde commented 3 days ago

Hello, author. I spent about 2 hours running the MVTec classification work on an RTX 4070. Is this time reasonable? The results under 1-shot are about 1 point lower (for MVTec) and 5 points lower (for VISA) compared to the data in your paper. Is this normal, or could there be some other issue?

FuNz-0 commented 16 hours ago

Please provide the hyperparameters for your running code

@.***

From: kirnde Date: 2024-10-18 22:50 To: FuNz-0/PromptAD CC: Subscribed Subject: [FuNz-0/PromptAD] Runtime and Results (Issue #21) Hello, author. I spent about 2 hours running the MVTec classification work on an RTX 4070. Is this time reasonable? The results under 1-shot are about 1 point lower (for MVTec) and 5 points lower (for VISA) compared to the data in your paper. Is this normal, or could there be some other issue? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>

kirnde commented 6 hours ago

I used the default settings from the original code, with 100 epochs. Today's experimental results show that the VISA results are similar to those in the paper, but the MVTec results are about 1-2 points lower. The results from three runs are exactly the same. Is the runtime for this experiment considered long? It mainly relies on the CPU, and the GPU consumption is not very high.