DPE codes for all 8 P1-P4 captures; note the P5C2/P6C2 gap

README.md

@@ -485,8 +485,16 @@ convex combination as augmentation.
 | P2C2 | 800 | cam07, cam30 | 200 |
 | P3C1 | 3664 | cam07, cam30 | 916 |
 | P3C2 | 814 | cam07, cam30 | 203 |
+| P4C1 | 3098 | cam07, cam30 | 774 |
+| P4C2 | 526 | cam07, cam30 | 131 |
 
-Remaining captures are being extracted and will appear here as they finish.
+Counts are codes, not frames: each frame contributes one code per camera, so a capture of
+*N* frames shot with two cameras yields 2*N*.
+
+**`P5C2` and `P6C2` have no DPE codes.** They are the cross-reenactment driving sequences,
+and only the eight `P1`..`P4` captures were extracted. If you want to drive an avatar with
+`P5C2` or `P6C2` on the DPE path, generate them yourself with `scripts/extract_dpe_codes.py`
+(the command below), or use the SMPL-X path, which needs nothing extra.
 
 **How they were made.** The code is `mlp_exp(dir(mlp(enc(face_crop))))` from
 [OpenTalker/DPE](https://github.com/OpenTalker/DPE)'s `Generator(size=256, style_dim=512,