search

analogdevicesinc / ai8x-synthesis

Quantization and Synthesis (Device Specific Code Generation) for ADI's MAX78000 and MAX78002 Edge AI Devices
Apache License 2.0
55 stars 47 forks source link

Test results with a pretrained Camvid #299

Closed kirilllzaitsev closed 11 months ago

kirilllzaitsev commented 1 year ago

Using the gen-demos-max78000.sh script to synthesize a Camvid model on MAX78000:

python ai8xize.py --test-dir $TARGET --prefix camvid_unet --checkpoint-file trained/ai85-camvid-unet-large-fakept-q.pth.tar --config-file networks/camvid-unet-large-fakept.yaml $COMMON_ARGS --overlap-data --mlator --no-unload --max-checklines 8192 --new-kernel-loader --overwrite "$@"

and SerialLoader.py from aisegment_unet-demo to test the inference pipeline, I arrive at these predictions: image displayed by:

ax[0].imshow(img_resize1)
ax[1].imshow(colors, cmap="Greys")
ax[2].imshow(img_resize1)
ax[2].imshow(colors, cmap="Greys", alpha=0.2)

While some masks look correct (the top part), I want to ask if stripped patterns and the quality of prediction, in general, are expected from a trained model trained/ai85-camvid-unet-large-fakept-q.pth.tar.

aniktash commented 1 year ago

The demo project for camvid is UNet-highres-demo which has 4 output classes for buildings, plants, sky, and unknown. Please use the SerialLoader.py in UNet-highres-demo\Utility if you would like to segment the scenery images. Figure_1

The aisegmentation_unet-demo is for the segmentation of portrait face images and has 2 classes for face and background. The SerialLoader in this project is slightly different. Figure_2

github-actions[bot] commented 11 months ago

This issue has been marked stale because it has been open for over 30 days with no activity. It will be closed automatically in 10 days unless a comment is added or the "Stale" label is removed.