Hf remover#1
Conversation
…_hff_pipeline.py.
eroux
left a comment
eroux
left a comment
There was a problem hiding this comment.
please remove masking code and see the comments
|
|
||
| | field | type | description | | ||
| |---|---|---| | ||
| | `w_id` | string | Work ID | |
There was a problem hiding this comment.
no need to have the w_id and i_id in the parquet files
|
|
||
| Each bbox is four `[x, y]` corner points (top-left → top-right → bottom-right → bottom-left). | ||
|
|
||
| Files are written to `<output-dir>/<w_id>/<i_id>/detections.jsonl` and `detections.parquet`. |
There was a problem hiding this comment.
no need for a jsonl, all the info is in the parquet
| ].read() | ||
|
|
||
| # ── detect ───────────────────────────────────────────────────── | ||
| bgr = _bytes_to_bgr(body) |
There was a problem hiding this comment.
this is quite fragile at the moment, there's a lot of images with very different characteristics on BDRC, please use better tested code like https://github.com/buda-base/bec-orchestration/blob/main/bec_orch/jobs/shared/decoder.py#L760 . Also please understand what frame size Surya operates on so the code can directly resize without having the Surya preprocessor do another resizing pass
|
|
||
| # ── mask (white-out HFF regions) ─────────────────────────────── | ||
| if dets: | ||
| result_bgr = _mask_hff_regions(bgr, dets, margin=self.cfg.margin) |
There was a problem hiding this comment.
as I tried to express many times, no masking please
| result_bgr = bgr | ||
|
|
||
| # ── upload ───────────────────────────────────────────────────── | ||
| out_bytes = _bgr_to_bytes( |
…ove unused parameters and simplify row construction.
…te JSONL file handling.
…me for improved frame decoding and remove unused _bytes_to_bgr function.
2 participants
detail