בלוג

בינה מאלכותית RB14-19 : הרצה של yolo על PASCAL VOC 2007

Rules for Training and Testing on PASCAL VOC 2007

1. If you train on other images (extra data), you must say so when you report your results.

   • Example: “Model trained on my custom dataset + VOC 2007, tested on VOC 2007 test set.”

   • This keeps your results honest and transparent.

2. Do not train on the VOC 2007 test set.

   • The test set is only for checking performance after training.

   • Using it for training makes your results invalid.

3. The main goal of testing on VOC 2007 is to check how well your model generalizes to the 20 official object classes.

   • You can train on other datasets, then test on VOC 2007 to measure how well your model works on standard benchmark images.

4. You can train your model on any dataset — not just PASCAL VOC.

5. You can test your model on the VOC 2007 test set — this is allowed.

6. If you train only on VOC 2007 (or VOC 2007+2012), your results are directly comparable to the official benchmark.

7. If you use outside data, your results are valid, but marked as “trained with extra data.”