Commit Graph
10 Commits
Author SHA1 Message Date
Kristian MyramandClaude Sonnet 5.5 7eb91ca44c S2.43: track several objects in one SAM 2 pass
TrackRequest takes a list of objects (each with its prompts on its frames) and
the tracker reports (object, frame, result); frame encodings are shared, objects
may start on different frames. VideoTracker::propagate_objects runs a subset.
Limits: 32 objects. Test: two discs, one prompted on frame 0 and one on frame 2.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-30 08:43:31 +01:00
Kristian MyramandClaude Sonnet 5.5 a62d59da4b S2.40: SAM 2 as the Roto Brush object tracker
ObjectTracker interface in the segment crate (prompts on some frames, the
clip tracked both ways, each frame streamed with confidence and an occluded
flag), implemented by SAM 2. MaskModel::tracker() is None for other models.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-30 08:43:30 +01:00
Kristian MyramandClaude Sonnet 5.5 92ac189a49 S2.31: load SAM 2 through the install flow; SAM 2 as a MaskModel
Sam2Mask wraps the image path behind the Roto Brush's MaskModel interface
(cached frame encoding, bilinear mask resize). The engine loads the verified
weights natively (not on wasm). Checked with the real file: install, notice,
load, use by Roto Brush 3.0, deselect.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-30 08:43:30 +01:00
Kristian MyramandClaude Sonnet 5.5 d1bff2baf6 S2.24: video tracking matches Sam2VideoModel on a synthetic clip
Click on frame 2 of a 7-frame clip with an occlusion, tracked forwards and
backwards: IoU 1.0000 on all 9 frame results, worst mask-logit error 0.0038,
object scores equal to 3 decimals.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-30 08:43:30 +01:00
Kristian MyramandClaude Sonnet 5.5 5b0cc57f87 S2.23: video tracker session
Per-object memory banks, prompts on any frame, forward and backward
propagation with streaming results, cancellation, object pointers with
temporal encoding, pruning of memories no later frame can use, and checked
input handling. Hostile-input and cancel tests.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-30 08:43:30 +01:00
Kristian MyramandClaude Sonnet 5.5 ea2899d077 S2.22: memory attention with rotary encoding matches PyTorch
Self and cross attention with axial 2-D RoPE (keys tiled over memory frames,
object-pointer tokens left unrotated). Max abs error 2.6e-5.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-30 08:43:30 +01:00
Kristian MyramandClaude Sonnet 5.5 e4f4fea0f2 S2.21: memory encoder matches PyTorch
Mask down-sampler, feature fuser and projection, sine position encoding,
occlusion embedding and bfloat16 rounding of stored memories. Max abs error
3.5e-6 on the memory features, 6e-8 on the position encoding.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-30 08:43:30 +01:00
Kristian MyramandClaude Sonnet 5.5 8a6ad6f5e9 S2.13: prompt encoder and two-way mask decoder match PyTorch
A click on the synthetic image: mask logits within 1.1e-4 (range 11.6), IoU
head 1e-6, object score 2e-5, prompt embeddings 1e-7.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-30 08:43:30 +01:00
Kristian MyramandClaude Sonnet 5.5 d2291926f0 S2.12: FPN neck and vision encoder match PyTorch
Max abs error 9.7e-6 on the features, 6e-8 on the sine position encodings.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-30 08:43:30 +01:00
Kristian MyramandClaude Sonnet 5.5 92b3a34732 S2.11: Hiera backbone matches PyTorch block by block
Weights loader with shape/finite validation (S2.10), patch embed, bicubic
positional embedding, window/global attention, query pooling. Reference test
(skips without weights): max abs error 2.6e-5 over all 12 blocks.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-30 08:43:30 +01:00