hunglc007
Yolov4 Evaluation returns an mAP of 0
I am having an issue with Yolov4. I downloaded the weights, and convert to pb. I create the dataset and do the preprocessing of the dataset. Then I run the command
python evaluate.py --weights ./checkpoints/yolov4-416
I then do
python remove_space.py
And
python main.py --output results_yolov4_tf
An I just get a total mAP of 0, I have attached the results txt file.
results.txt
Is this a known problem, am I missing a step or do I need to change something?
Any help would be appreciated, thanks!
Tensorflow version: 2.6.0
Your issue may happen during converting the weights. In savemodel.py, go to the 51st row and replace utils.load_weights(model, FLAGS.weights, FLAGS.model, FLAGS.tiny) with model.load_weights(FLAGS.weights) . Then redo what the demo.
Thanks for the speedy reply!
I get an issue from this line:
OSError: Unable to open file (file signature not found)
I have tried redownloading the weights file but I get the same error.
Edit: My colleague says he managed to do the workflow with this amend but says he got the same result (mAP of 0), I will verify with him as to why I got an error and he didnt.
Rolling back your tf version as decribed in requirements-gpu.txt may be helpful.
I met the same mAP issue before. However ,my issue only came out when I used the weights that was automatically saved during training instead of the default weights. After my correction as above ,the mAP issue was solved. However, it was still poor(about 42%). But it's another thing.
facing the same issue with yolov4-tiny, any updates on this. although the tflite detections give good results.
