Add scripts for models evaluation on image sequences#1
Add scripts for models evaluation on image sequences#1
Conversation
MateoLostanlen
left a comment
MateoLostanlen
left a comment
There was a problem hiding this comment.
Hi @theocayla , thanks for the pr . I put 2 small comments that need to be corrected but nothing bad it looks generally ok. I'm thinking it might be interesting to have stats by sequence too, but we can do that later.
| json.dump(existing_config, fp) | ||
|
|
||
| # Csv file where detailed predictions are dumped | ||
| outCsv = os.path.join(outpath, f"results/results_{configId}.csv") |
There was a problem hiding this comment.
you need to create this results folder automatically using os.makedirs for exemple, the code fail otherwise
| }, | ||
| } | ||
|
|
||
| run_engine_predictions(image_folder, datapath, config, savePred=True) |
There was a problem hiding this comment.
savePred do not exist in your function definition
|
Hey @MateoLostanlen thanks for the comments, I did the changes. Regarding your comment, what do you mean by "stats by sequence" ? Each sequence results is stored in the results dict under the key "details", which allows to track whether at least one detection was made among the sequence, do you think we should add the detection status of each image in the sequence ? |
2 participants
The aim is to call the model via the Pyro Engine on image sequences to simulate the complete processing pipe and provide metrics closer to real-life conditions.