Keras-yolo3: How to predict a test set of images?

[canyilmaz90]

Using yolo_video.py we can get results real-time, but is there any function to evaluate our model on thousands of images? I tried to read codes, unfortunately I couldn't find any function including something like model.predict() or model.evaluate().

If someone knows something about this, please let me know.

[bluesy7585]

I had same question. but couldn't find the answer.
so I made some changes to the code. it process images in folder and output txt files.
python yolo_video.py --imgdir .\sample_folder --txt
https://github.com/bluesy7585/keras-yolo3

[canyilmaz90]

I had same question. but couldn't find the answer.
so I made some changes to the code. it process images in folder and output txt files.
python yolo_video.py --imgdir .\sample_folder --txt
https://github.com/bluesy7585/keras-yolo3

@bluesy7585 thank you for your reply. Have you modified only the "yolo.py" and "yolo_video.py" scripts?

[bluesy7585]

@canyilmaz90
I mode some changes in other files as well, but it only effects training phase.

[canyilmaz90]

@bluesy7585 this worked very well for my case, thanks a lot!

[cuimiao187561]

How to evaluate the test set, if you have already solved it, can you send the code to me?([email protected]) @canyilmaz90

[canyilmaz90]

@cuimiao187561 look at the solution proposed by @bluesy7585 , codes are available in his/her repo and its very easy to implement.

[gzz1529657064]

I had same question. but couldn't find the answer.
so I made some changes to the code. it process images in folder and output txt files.
python yolo_video.py --imgdir .\sample_folder --txt
https://github.com/bluesy7585/keras-yolo3

I read your code.But after using your code, executing the command will still give an error. In my yolo_video.py, I can detect video.
"python yolo_video.py --imgdir ./detection_results/test_image/ --txt"
./detection_results/test_image/ is my test image folder
executing command error!
This is my WeChat: wxid_cbjyy2ydd0j422 May I ask you some questions? Thanks!