Datasets:
|
Download KNOWN_PITFALLS.md from simplexsigil2/omnifall: direct link, hf CLI and curl.
- Browser
- Download file 2.23 kB
-
https://huggingface.co/datasets/simplexsigil2/omnifall/resolve/main/KNOWN_PITFALLS.md
- Command line
-
hf download hf://datasets/simplexsigil2/omnifall/KNOWN_PITFALLS.md
-
curl -L -o KNOWN_PITFALLS.md https://huggingface.co/datasets/simplexsigil2/omnifall/resolve/main/KNOWN_PITFALLS.md
2.23 kB
| # Known Pitfalls | |
| In this file we document known problems and pitfalls. | |
| --- | |
| ## 1. In the multi-view components, some segments end after the video does | |
| **What you will see.** In `edf`, `cmdfall` and `mcfd`, a segment whose `end` lies | |
| past the last frame of its video. Depending on your decoder you get fewer frames | |
| than you asked for, or an error. | |
| | Component | Segments affected | Videos affected | Largest overrun | | |
| |---|---|---|---| | |
| | mcfd | 85 / 1,352 | 71 / 192 | 1.19 s (`chute05/cam8`) | | |
| | edf | 18 / 508 | 5 / 10 | 7.36 s (`jinhui/view2/rgb/...`) | | |
| | cmdfall | 76 / 42,143 | 75 / 384 | 3.80 s (`colors/S22P36K6`) | | |
| **Why.** These components are multi-camera, and one view was annotated, afterwards the labels were propagated to other views. | |
| To make sure the labels are still correct, each dataset carries per-camera synchronisation offsets; a | |
| secondary view's segments are the reference view's segments shifted by that | |
| offset. Where the offset is positive, or the secondary camera stopped recording | |
| earlier than the reference, the shifted tail segment lands beyond the end of the video file. | |
| --- | |
| ## 2. Some config names are kept aliases | |
| Config names have been revised, and the earlier spellings were kept working | |
| rather than removed, so that code written against them does not break. Both | |
| names load the same data. | |
| | Old name | Current name | | |
| |---|---| | |
| | `cs-staged`, `cv-staged` | `of-sta-cs`, `of-sta-cv` | | |
| | `cs-staged-wild`, `cv-staged-wild` | `of-sta-to-all-cs`, `of-sta-to-all-cv` | | |
| | `of-sta-to-itw-*`, `of-sta-itw-*` | `of-sta-to-all-*` | | |
| | `of-syn-to-itw`, `of-syn-itw` | `of-syn-to-all-cs` | | |
| | `of-sta-syn-to-itw-*` | `of-sta-syn-to-all-*` | | |
| | `OOPS` | `of-itw` | | |
| | `caucafall`, `cmdfall`, `edf`, `gmdcsa24`, `le2i`, `mcfd`, `occu`, `up_fall` | the matching `-cs` config | | |
| Two consequences worth knowing. A bare component name such as `le2i` is an alias | |
| for `le2i-cs`, so it gives you the cross-subject split rather than "all of le2i" | |
| --- if you want the cross-view split, ask for `le2i-cv` explicitly. And when you | |
| report which configuration you trained on, prefer the current name, since the | |
| aliases are the ones most likely to be retired later. | |
| [CONFIGS.md](CONFIGS.md) carries the full list. | |