"a penguin"
frame t
Detector
masks detected in frame t
masks propagated
from frame t-1
merge existing and
newly detected masks
output for frame t
Tracker
exemplar
RepViT-M0.9
RepViT-M1.1
RepViT-M2.3
TinyViT-5M
TinyViT-11M
TinyViT-21M
EfficientViT-B0
EfficientViT-B1
EfficientViT-B2
4.72M
7.77M
22.40M
5.07M
10.55M
20.62M
0.68M
4.64M
14.98M
Image
Encoder
Variant of your choice
from SAM3
from MobileCLIP
Trainig Data
M0.9
M1.1
M2.3
 5M 
 11M
 21M
B0
B1
B2
Model Card
Distill
Architecture of your choice
Memory
Bank
Efficient
Memory Bank
Replace
Perciever
Insert
from EfficientTAM
from EdgeTAM
Efficient
SAM3
RepViT
TinyViT
EfficeintViT
mask, point
   or box
Text
Encoder
Distill
mobileclip
["a penguin", "sea"]
S0
S1
L
Text Encoder of your choice
MobileCLIP-S0
42.57M
MobileCLIP-S1
63.56M
MobileCLIP-L
123.6M
from Efficient Models