FFCV Imagenet Image Quality vs. Native Imagenet Image Quality

Hello! I see that FFCV offers alot of options for the quality of the dataset - e.g. in the imagenet example:

# - 50% JPEG encoded, 90% raw pixel values
# - quality=90 JPEGs
./write_dataset.sh 500 0.50 90

One thing I’m curious is the effects of these quality options on the training results of the dataset, as I’m interested in reproducing Imagenet results but faster using FFCV. What would also be recommended settings to use if you would like to produce Native Imagenet Quality precisely (barring the crop size).

[andrewilyas]

Just to add: for our ImageNet results (the speed/accuracy tradeoff scatterplot and table), we use:

Max image size: 500
Random compression with 50% probability
JPEG Quality 90

[GuillaumeLeclerc]

I think this thread has been dealt with. @YassineYousfi Feel free to start a new conversation about these specific topics.