Serve MMDetection
Deploy model as REST API service
Deploy model as REST API service
Overview • Pretrained models • How To Run • Related Apps • Acknowledgment
Serve Detectron2 (Instance Segmentation) model as Supervisely Application.
Model serving allows to apply model to image (URL, local file, Supervisely image id) with 2 modes (full image, image ROI). Also app sources can be used as example how to use downloaded model weights outside Supervisely.
Application key points:
Detectron2 provides us Mask R-CNN Instance Segmentation baselines based on 3 different backbone combinations:
We have integrated popular architectures into this application.
ℹ️ You can find more information about each model here (use model id)
model | train time (s/im) | inference time (s/im) | box AP | mask AP | model id |
---|---|---|---|---|---|
R50-C4 (1x) | 0.584 | 0.11 | 36.8 | 32.2 | 137259246 |
R50-DC5 (3x) | 0.47 | 0.076 | 40 | 35.9 | 137849551 |
R50-FPN (100) | 0.376 | 0.069 | 44.6 | 40.3 | 42047764 |
R50-FPN (400) | 0.376 | 0.069 | 47.4 | 42.5 | 42019571 |
R101-FPN (100) | 0.376 | 0.069 | 46.4 | 41.6 | 42025812 |
R101-FPN (400) | 0.376 | 0.069 | 48.9 | 43.7 | 42073830 |
regnetx4gfdds_FPN (100) | 0.474 | 0.071 | 46 | 41.3 | 42047771 |
regnetx4gfdds_FPN (400) | 0.474 | 0.071 | 48.6 | 43.5 | 42025447 |
regnety4gfdds_FPN (100) | 0.487 | 0.073 | 46.1 | 41.6 | 42047784 |
regnety4gfdds_FPN (400) | 0.487 | 0.073 | 48.2 | 43.3 | 42045954 |
model | train time (s/im) | inference time (s/im) | box AP | mask AP | model id |
---|---|---|---|---|---|
R50-FPN | 0.292 | 0.107 | 23.6 | 24.4 | 144219072 |
R101-FPN | 0.371 | 0.114 | 25.6 | 25.9 | 144219035 |
X101-FPN | 0.712 | 0.151 | 26.7 | 27.1 | 144219108 |
model | train time (s/im) | inference time (s/im) | box AP | mask AP | model id |
---|---|---|---|---|---|
R50-FPN | 0.24 | 0.078 | - | 36.5 | 142423278 |
model | train time (s/im) | inference time (s/im) | box AP | mask AP | model id |
---|---|---|---|---|---|
Deformable Conv (3x) | 0.349 | 0.047 | 42.7 | 38.5 | 144998336 |
Cascade R-CNN (3x) | 0.328 | 0.053 | 44.3 | 38.5 | 144998488 |
GN (3x) | 0.309 | 0.06 | 42.6 | 38.6 | 138602888 |
Mask R-CNN X152 | - | 0.234 | 50.2 | 44 | 18131413 |
You can use served model in next Supervisely Applications ⬇️
Apply
button (or use hotkey) and detections with their confidences will immediately appear on the image. This app is based on the great work Detectron2
(github).
Deploy model as REST API service
Deploy MMDetection 3.0 model as a REST API service
Deploy Transfiner for instance segmentation as a REST API service
Training dashboard for mmdetection framework (v3.0.0 and above).
Dashboard to configure, start and monitor training