search

mikeyEcology / MLWIC

Machine Learning for Wildlife Image Classification
70 stars 16 forks source link

accuracy when using model for your images #24

Open mikeyEcology opened 5 years ago

mikeyEcology commented 5 years ago

Have you used MLWIC's classify function on your images? It seems like a lot of users have found low accuracy I'd like to collect some data on this. A common problem in these types of models is transferability (the ability of a model to work on out-of-sample images), and I want to do a follow up analysis on the (non) transferability of the built in model in MLWIC. If you have used MLWIC, I'm hoping you'd be willing to share some data on how the model worked on your images.

I'm looking for the following information and if you could provide the information in a table (csv) with the following columns that would be best: Species, number true positives (number of times the model was correct in classifying this species as the top guess), false negatives (number of times this species was in the image, but something else was the top guess), false positives (number of times the model said this species was in the image but it is not actually there), location (state, county, zip code, ... as much detail as you are willing and able to provide), other information about your study (if applicable).

If you are willing to share, please email the information to tabakma [at] gmail or post it as a response in this issue if it is easier for you.

Nova-Scotia commented 5 years ago

Hi @mikeyEcology , I feel like at the current time my estimates aren't reliable (as I have issues with images manually tagged by "photo series", not event), but I'll keep you posted once we've got a cleaner dataset for testing (in the works).

Thank you for continuing to support and investigate this science!!!