Hi all,
My UiPath image classification is not working properly, i am using 10 images in each class (cats,bus,bikes) in training the data and and 4 images in each class to evaluate the data
Hi all,
My UiPath image classification is not working properly, i am using 10 images in each class (cats,bus,bikes) in training the data and and 4 images in each class to evaluate the data
also please let me know do i have to use the same images in evaluate that i used in train? this is a question i really want to know
Hi @Jai_Pande , no, dont use the same images for evaluation. You should use different images to know how well the model have been trained.
Hi @Jai_Pande ,
Could you maybe provide us with the dataset used for Training / Evaluation so that we could also maybe check it from our end ?
sure okay got it, and in the predictions.csv the file name it shows its from train or evaluate
ps:- also spelling of predictions is wrong (just a minor quality feedback)
@supermanPunch
Data.zip (341.6 KB)
anything brother???
Is this the Predictions.csv file ?
Because After the Full Pipeline run with the dataset you have provided, i was able to get the proper evaluation in the predictions.csv file :
yes it was the predications.csv only brother… and i used train and evaluate pipelines instead of full
Maybe the Version selected was 0 instead of 1 (Considering this is the highest version with High trained dataset) for Evaluation ?
i made sure of it brother… and cant i use train and evaluate or its mandatory that i have to use full only
When checked the same way - First Train Run and then Evaluation Run, the predictions are not correct.
Maybe we would require to train with more similar images (a larger dataset for each class) and then check with it.
okay got it i will do it, also can you kindly let me know the workflow after model has classified the images, for e.g the confidence score is less than 0.4 then we need to create a form task and after the human validation we can upload the file back to the mode?
Could you create a new Topic for the Form Task related queries, as we could separate the Initial Query of the ML Model behaviour solely for this topic and the discussions on the Form Task could happen on a different Thread.
You can keep this topic open as the Behaviour of the model is not clarified completely yet.
Looks like the new images might be little different…so please train more files and then check
cheers
i have texted you regarding a workflow on your insta account