pjm-models / README.md
zamog's picture
Add files using upload-large-folder tool
f20a8cc verified
|
Raw
History Blame Contribute Delete
2.57 kB
# PJM Sign-Language β†’ English β€” ONNX models
Repo-free ONNX exports of two PJM (Polish Sign Language) β†’ English translation model families,
served by the `signsdk` runtime (onnxruntime + bundled tokenizers/assets β€” no model-repo needed).
All graphs **fp32**. Each model = `projector.onnx` + `encoder.onnx` + `decoder.onnx` (+ `.onnx.data`
sidecar for SpaMo's flan-t5-xl) + `meta.json` + prompt/prefix embedding.
- **Uni-Sign** β€” pose-based (RTMPose keypoints) β†’ mT5. Self-contained graphs.
- **SpaMo / Mod-SpaMo** β€” CLIP ViT-L/14 + VideoMAE-large (+ hand-CLIP + MediaPipe for `mod-`) β†’ flan-t5-xl.
Shares `spamo/shared/` (CLIP+VideoMAE+hand-CLIP) across all SpaMo checkpoints.
## Layout
```
assets/ mt5_tokenizer/ flan_t5_tokenizer/ yolo/yolo11n.onnx mediapipe/*.task
unisign/<name>/ projector.onnx encoder.onnx decoder.onnx prefix.npy meta.json
spamo/shared/ clip_vit_s2.onnx(+.data) videomae.onnx(+.data) hand_clip.onnx(+.data)
spamo/<name>/ projector.onnx(+.data) encoder.onnx(+.data) decoder.onnx(+.data) prompt_embeds.npy meta.json
```
## Models (test-set BLEU-4, βœ“ = uploaded)
| model | family | source pretrain | split | BLEU-4 | status |
|-------|--------|-----------------|-------|--------|--------|
| `unisign-pjm-cslnews-ms` | Uni-Sign | CSL-News | ms | 2.42 | βœ“ |
| `unisign-pjm-cslnews-si` | Uni-Sign | CSL-News | si | 1.43 | βœ“ |
| `unisign-pjm-how2sign-ms` | Uni-Sign | How2Sign | ms | 7.61 | βœ“ |
| `unisign-pjm-how2sign-si` | Uni-Sign | How2Sign | si | 4.77 | β€” |
| `unisign-pjm-openasl-ms` | Uni-Sign | OpenASL | ms | 7.38 | β€” |
| `unisign-pjm-openasl-si` | Uni-Sign | OpenASL | si | 4.79 | β€” |
| `unisign-pjm-scratch-ms` | Uni-Sign | scratch | ms | 1.62 | β€” |
| `unisign-pjm-scratch-si` | Uni-Sign | scratch | si | 0.72 | β€” |
| `spamo-pjm-phoenix-ms` | SpaMo | PHOENIX | ms | 5.54 | βœ“ |
| `spamo-pjm-phoenix-si` | SpaMo | PHOENIX | si | 0.53 | β€” |
| `spamo-pjm-scratch-ms` | SpaMo | scratch | ms | 2.03 | β€” |
| `spamo-pjm-scratch-si` | SpaMo | scratch | si | 0.93 | β€” |
| `mod-spamo-pjm-phoenix-ms` | Mod-SpaMo | PHOENIX | ms | 5.06 | βœ“ |
| `mod-spamo-pjm-phoenix-si` | Mod-SpaMo | PHOENIX | si | 0.47 | β€” |
| `mod-spamo-pjm-scratch-ms` | Mod-SpaMo | scratch | ms | 1.56 | β€” |
| `mod-spamo-pjm-scratch-si` | Mod-SpaMo | scratch | si | 1.17 | β€” |
BLEU-4 = reported PJM test score. Decoding faithful to the original eval
(Uni-Sign beam=4; SpaMo beam-sample num_beams=5, top_p=0.9). fp32 chosen for deploy
(fp16 overflows the bf16-trained flan-t5-xl; int8 is size-only).