S2.50: bench --roto compares classical, MobileSAM and SAM 2

--mobilesam FILE; results recorded in docs/roto-brush-sam2.md (re-entry clip:
0.632 classical, 0.499 MobileSAM, 0.975 SAM 2).

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This commit is contained in:
Kristian MyramandClaude Sonnet 5.5 committed 2026-10-30 08:43:30 +01:00
1 parent 371e67f51e
commit b72637bb55
2 files changed
+22 -17

No files matched your search

+9 -8
View File
@@ -27,7 +27,7 @@
//! (--gpu: CPU vs GPU ms/frame for every comp at Full and Half)
//! effectcraft-cli bench --ops [--small] [--layers N] [--comps N] [--footage N] everyday-operation timings
//! on a large generated project (open, save, auto-save, undo/redo, timeline, Project panel)
//! effectcraft-cli bench --roto [--frames N] [--sam2 model.safetensors] [--json] Roto Brush trackers on synthetic clips with known mattes
//! effectcraft-cli bench --roto [--frames N] [--sam2 model.safetensors] [--mobilesam mobile_sam.pt] [--json] Roto Brush trackers on synthetic clips with known mattes
//! (classical engine, MobileSAM and SAM 2.1 Tiny when installed; IoU, flicker, re-appearance, seconds per frame)
//! effectcraft-cli script FILE.jsx [F.ecproj] | --eval CODE run an After Effects-style script
//! effectcraft-cli mcp [--autosave | --bridge PORT] MCP server on stdio
@@ -126,7 +126,7 @@ const USAGE: &str = "usage: effectcraft-cli <info|commands|exec|run|props|get|se
bench --ops [--small] [--layers N] [--comps N] [--footage N]
everyday operations on a large generated project: startup, open,
save, auto-save, edits + undo/redo, timeline, Project panel
bench --roto [--frames N] [--sam2 FILE] [--json]
bench --roto [--frames N] [--sam2 FILE] [--mobilesam FILE] [--json]
Roto Brush trackers on synthetic clips with known mattes: movement,
lighting change, occlusion, leaving and re-entering the frame
script FILE.jsx [F.ecproj] | --eval CODE run JavaScript with the After Effects-style object model
@@ -144,6 +144,7 @@ const VALUED: &[&str] = &[
"--n",
"--frames",
"--sam2",
"--mobilesam",
"--play",
"--layers",
"--comps",
@@ -1015,13 +1016,13 @@ fn bench_roto(args: &Args) -> Result<(), Failure> {
use effectcraft_track::roto::bench;
let frames = args.num("--frames")?.unwrap_or(24.0).clamp(12.0, 240.0) as usize;
let session = effectcraft_host::session();
// `--mobilesam FILE` uses the weights at FILE; otherwise the installed ones.
let mobilesam_file = args.opt("--mobilesam").map(std::path::PathBuf::from).or_else(|| session.model_path(&MOBILE_SAM));
let mobilesam: Option<std::sync::Arc<dyn MaskModel>> =
session.model_path(&MOBILE_SAM).and_then(|p| std::fs::read(p).ok()).and_then(|bytes| effectcraft_segment::load(MOBILE_SAM.id, &bytes).ok()).and_then(
|m| match m {
Loaded::Mask(m) => Some(m),
_ => None,
},
);
mobilesam_file.and_then(|p| std::fs::read(p).ok()).and_then(|bytes| effectcraft_segment::load(MOBILE_SAM.id, &bytes).ok()).and_then(|m| match m {
Loaded::Mask(m) => Some(m),
_ => None,
});
if mobilesam.is_none() {
note!("bench --roto: MobileSAM is not installed; comparing the classical engine only");
}
+13 -9
View File
@@ -44,14 +44,18 @@ clip costs one encode per frame, so a 300-frame clip takes on the order of 17 mi
backend; GPU work is Phase 6 of the plan (the early Hiera blocks are memory-bound and have headroom on the CPU too).
## Benchmark
`effectcraft-cli bench --roto [--frames N] [--sam2 model.safetensors]` runs synthetic clips with known mattes. 12 frames,
debug build, mean IoU (after the object returns):
`effectcraft-cli bench --roto [--frames N] [--sam2 model.safetensors] [--mobilesam mobile_sam.pt]` runs synthetic clips
with known mattes through all three. 12 frames, debug build, this PC; mean IoU (mean IoU after the object returns), seconds
per frame:
| Clip | Classical | SAM 2.1 Tiny |
|---|---|---|
| move | 0.982 | 0.993 |
| light | 0.981 | 0.992 |
| occlusion | 0.978 | 0.988 |
| reenter | 0.632 (0.000) | 0.975 (0.967) |
| Clip | Classical | MobileSAM | SAM 2.1 Tiny |
|---|---|---|---|
| move | 0.982 | 0.992 | 0.993 |
| light | 0.981 | 0.992 | 0.992 |
| occlusion | 0.978 | 0.881 | 0.988 |
| reenter | 0.632 (0.000) | 0.499 (0.725) | 0.975 (0.967) |
| s/frame | 0.01 | 1.1 | 8.2 |
MobileSAM's column appears when it is installed.
MobileSAM prompts every frame independently from the flow-warped matte, so it follows the object while the flow does and
loses it when it is hidden; SAM 2's streaming memory is what lets it re-acquire the object. The synthetic clips are easy
(one textured disc), so the "move" and "light" rows mostly show all three working; the re-entry row is the difference.