Pipeline Architecture
Explore the modular stages of the SATFLOW AI framework. Select a pipeline stage to inspect algorithms and references.
SATFLOW AI Process Pipeline
RIFE Flow EngineFlow Estimation (IFNet)
Real-time Intermediate Flow Estimation (RIFE). Leverages a multi-scale CNN (IFNet) to estimate intermediate flow vectors from T0 and T1 directly, avoiding complex independent optical flow calculations.
Operational Details
- Estimates coarse-to-fine optical flow fields at 3 resolution scales.
- Computes soft fusion occlusion masks to identify overlapping/disappearing clouds.
- Accepts arbitrary timesteps t ∈ [0, 1] to synthesize any intermediate temporal position.
Mathematical Formula
I_t = (1-t) * W(I_0, F_t->0) + t * W(I_1, F_t->1)
F_t->0 = t * F_1->0, F_t->1 = (1-t) * F_0->1
Python Source Implementation
# Import official RIFE class from cloned module
from train_log.RIFE_HDv3 import Model
model = Model()
model.load_model('train_log')
model.eval()
# Interpolate at arbitrary timestep t (e.g. 0.5)
with torch.no_grad():
output_tensor = model.inference(img0_tensor, img1_tensor, scale=1.0)