balancap / SSD-Tensorflow

Single Shot MultiBox Detector in TensorFlow
4.1k stars 1.89k forks source link

mAP is 0 on my custom dataset #129

Open gul98 opened 6 years ago

gul98 commented 6 years ago

I have trained my own model on my own dataset. It contains only one class. But when i run evaluation it shows 0 accuracy. Any one else ran into this problem?

congjianting commented 6 years ago

have you fined the solution? i face the same problem.

liuguiyangnwpu commented 6 years ago

how to plot the auc curve, and how to calc the FP TP to calc F-Value???? when i use the eval_ssd_network.py, I use tensorboard to show the result. No curve in it !

Cuky88 commented 6 years ago

Did you find a solution?

kritchie commented 6 years ago

Hi guys, I had the same issue yesterday, mAP of 0. Couple things you should check :

I hope it can help some of you !

Cheers !

rainley123 commented 5 years ago

Hi guys, I had the same issue yesterday, mAP of 0. Couple things you should check :

* when you use the checkpoint : "--checkpoint_path=/path/to/checkpoint.ckpt" be sure to have unzipped the files correctly. Meaning you should not have a folder named checkpoint.ckpt (this is just an arbitrary example, like "ssd_300_vgg.ckpt"), but only 2 files checkpoint.ckpt.data-00000-of-00001 and checkpoint.ckpt.index. Be sure to remove any folder that resulted in the extraction process.

* I was using thermal infrared images for detection. Without retraining the model could not detect objects. I found that out by using the Notebook provided and added my images in the "demo" and changed the demo image displayed. That way I could find out the model is working correctly but just couldn't find objects.

I hope it can help some of you !

Cheers !

hello, I don't use all these codes in the respository, but some core codes, I have the same problem that the accuracy is nearly 0, How can I solve it