ModelVerificadoExterno

Maestro Lstm

TipoOtherPublicador@ntsrigaudCnota

Fonte do modelo

Tipo: Other
Tamanho do arquivo: 549 KB–13,9 MB
Atualizado: 14 de set. de 2026
Publicador: @ntsrigaud

Trecho da fonte

Artefatos de modelo

35 artefatos

lstm_v1_20260418T004149Z_inference.pt

pt · 1,05 MB · SHA-256 b549b45f9275…94b1

lstm_v1_20260418T023120Z_inference.pt

pt · 1,05 MB · SHA-256 5c8bfaed0908…6800

two_stream_attn_v1_20260503T061857Z_inference.pt

pt · 4,63 MB · SHA-256 4607db63f5a7…378c

two_stream_attn_v1_20260503T065204Z_inference.pt

pt · 4,63 MB · SHA-256 e125b5db868b…e4a4

two_stream_attn_v1_20260504T131754Z_inference.pt

pt · 8,03 MB · SHA-256 3c16988d40a9…25d6

two_stream_attn_v1_20260504T222832Z_inference.pt

pt · 8,03 MB · SHA-256 dc57b6accd65…2627

two_stream_attn_v1_20260504T223608Z_inference.pt

pt · 11,4 MB · SHA-256 b92d06530ca6…8986

two_stream_attn_v1_20260504T225032Z_inference.pt

pt · 8,03 MB · SHA-256 f46f52679d24…0995

two_stream_attn_v1_20260504T230000Z_inference.pt

pt · 8,03 MB · SHA-256 e66c5231bdc0…b957

two_stream_attn_v1_20260504T233549Z_inference.pt

pt · 8,03 MB · SHA-256 e57a6bd35eea…fd1f

two_stream_attn_v1_20260504T234516Z_inference.pt

pt · 8,03 MB · SHA-256 222d436e12fe…f7d0

two_stream_attn_v1_20260505T011026Z_inference.pt

pt · 8,03 MB · SHA-256 eaa044de420b…1677

two_stream_attn_v1_20260505T012223Z_inference.pt

pt · 8,03 MB · SHA-256 011e431bdcab…b601

two_stream_attn_v1_20260505T094024Z_inference.pt

pt · 8,03 MB · SHA-256 9ed71ba9ad1d…b9c8

two_stream_attn_v1_20260505T142512Z_inference.pt

pt · 7,90 MB · SHA-256 c7b9e1c8a5c9…e7b9

two_stream_attn_v1_20260505T224138Z_inference.pt

pt · 7,90 MB · SHA-256 8a8d6831a72d…b47e

two_stream_attn_v1_20260505T225339Z_inference.pt

pt · 13,9 MB · SHA-256 cad2e2176b38…3c8b

two_stream_attn_v1_20260505T225847Z_inference.pt

pt · 7,90 MB · SHA-256 5b77b4c55ad3…bae1

two_stream_attn_v1_20260506T014550Z_inference.pt

pt · 7,90 MB · SHA-256 f96b854510ca…3866

two_stream_attn_v1_20260506T215509Z_inference.pt

pt · 7,90 MB · SHA-256 e2e54043a58f…a5fc

two_stream_attn_v1_finetune_20260510T062759Z_inference.pt

pt · 8,03 MB · SHA-256 b5e475e4ed0d…dddd

two_stream_attn_v1_finetune_20260510T072119Z_inference.pt

pt · 8,03 MB · SHA-256 4c2dcca1bd7b…fe69

two_stream_attn_v1_finetune_20260510T081052Z_inference.pt

pt · 8,03 MB · SHA-256 c61720f88215…c9d0

two_stream_attn_v1_finetune_20260510T082713Z_inference.pt

pt · 8,03 MB · SHA-256 b8e6e461c00d…a94e

two_stream_attn_v1_finetune_20260510T085841Z_inference.pt

pt · 8,03 MB · SHA-256 94b0570445c8…ba2e

two_stream_attn_v1_finetune_20260510T094220Z_inference.pt

pt · 8,03 MB · SHA-256 524badee81d1…4726

two_stream_attn_v1_finetune_20260510T095606Z_inference.pt

pt · 8,03 MB · SHA-256 e81ddd590813…a3bb

two_stream_attn_v1_finetune_20260510T132307Z_inference.pt

pt · 8,03 MB · SHA-256 809f3509bbab…04d8

two_stream_attn_v1_finetune_20260510T135744Z_inference.pt

pt · 8,03 MB · SHA-256 e16a8b4900b7…a760

two_stream_attn_v1_finetune_20260510T160508Z_inference.pt

pt · 8,03 MB · SHA-256 ebc4989d677b…81ca

two_stream_lstm_v1_20260420T091431Z_inference.pt

pt · 549 KB · SHA-256 8213c8c17d3b…6c72

two_stream_lstm_v1_20260420T091839Z_inference.pt

pt · 549 KB · SHA-256 f82efd4f8e94…d1d4

two_stream_lstm_v1_20260422T124425Z_inference.pt

pt · 549 KB · SHA-256 7b6d236ffb59…9860

two_stream_lstm_v1_20260422T125010Z_inference.pt

pt · 549 KB · SHA-256 f002c012417b…3d01

two_stream_lstm_v1_20260422T130834Z_inference.pt

pt · 549 KB · SHA-256 b1b994d6388f…4cca

Trechos de fonte

3 trechos

Readme11 de jul.huggingface.co/ntsrigaud/maestro-lstm

--- language: - en license: mit tags: - gesture-recognition - hand-gesture - pytorch - mediapipe - temporal-model - lstm - attention - bidirectional datasets: - IPN-Hand metrics: - accuracy - f1 model-index: - name: two_stream_attn_v1_finetune_20260510T160508Z results: - task: type: gesture-recognition dataset: name: IPN Hand type: IPN-Hand metrics: - type: accuracy value: 0.8623 - type: f1 value: 0.8577 --- # two_stream_attn_v1_finetune_20260510T160508Z A real-time hand gesture classifier trained on the [IPN Hand dataset](https://gibranbenitez.github.io/IPN_Hand/). This model is part of the **Maestro** pipeline that enables touchless control of presentation and meeting software through hand gestures captured from a standard webcam using MediaPipe for landmark extraction. ## Model Description - **Architecture**: EnhancedTwoStreamLSTM (BiLSTM h=128×2, MHA 8 heads, proj=96, mean+max pool, MLP gate) - **Parameters**: 2,099,434 - **Input**: `(batch, 32, 147)` — 32-frame sliding window at 30 FPS ≈ 1067 ms - **Output**: Softmax logits over 10 gesture classes - **Inference latency**: < 1 ms per call (CPU, single sample) - **Feature schema**: `feature-schema-v5` ## Architecture `EnhancedTwoStreamLSTM` splits the 147-dim feature vector into two parallel streams and processes them through a BiLSTM + self-attention + MLP-gate pipeline: ``` Input (B, T=32, 147) │ ├─ Stream A — Pose/Shape (73 dims) │ Linear+LN+GELU → 96 │ 2-layer BiLSTM (h=128) → (B, T, 256) │ LayerNorm → Self-MHA (8 heads) + residual + post-LN │ mean+max pool → pool_LN → ctx_a (B, 256) │ ├─ Stream B — Motion/Dynamics (74 dims) │ (identical structure) → ctx_b (B, 256) │ ├─ MLP cross-stream gate │ gate_a = Sigmoid( │ Linear(128→256)( │ Tanh(Linear(256→128)(ctx_b)))) │ ctx_a = LN(ctx_a × gate_a + ctx_a) │ gate_b = Sigmoid( │ Linear(128→256)( │ Tanh(Linear(256→128)(ctx_a)))) │ ctx_b = LN(ctx_b × gate_b + ctx_b) │ └─ cat(ctx_a, ctx_b) → (512,) LN → Linear(512→256) → GELU → Dropout → Linear(256→10) ``` **Design rationale:** - BiLSTMs encode temporal order via their recurrent cell state — no positional encoding needed. - Mean+Max pooling captures both sustained gesture shape (mean) and transient click events (max). - The 2-layer MLP gate provides non-linear cross-modal recalibration at ~37 K params (vs ~263 K for full MHA cross-attention with a degenerate mean-poo...

Note11 de jul.huggingface.co/ntsrigaud/maestro-lstm

Source context: 21 downloads · 0 likes · Repo ntsrigaud/maestro-lstm

Note11 de mai.huggingface.co/api/models/ntsrigaud/maestro-lstm

Source context: 543 downloads · 0 likes · Repo ntsrigaud/maestro-lstm