Mike

Bbox results after tranformations are off

11

## 🐛 Bug ## To Reproduce I load the same image with my training and validation augmentation stacks. My val augmentation stack: ```python self.val_transforms = A.Compose( [ # LETTERBOX WITH...

Running [this](https://github.com/cheind/py-motmetrics#for-custom-dataset) example gives different results than running `motmetrics.apps.eval_motchallenge`: ``` IDF1 MOTA acc 0.71134 0.552 ``` vs ``` IDF1 MOTA acc 73.2 53.7 ``` Their default configuration seems to be...

Any plans of creating a pip package out of this repo?

3

?

`fuse_score` vs `fuse_motion`

2

Why is it that `fuse_motion`, i.e.: ```python raw_emb_dists = matching.embedding_distance(strack_pool, detections) dists = matching.fuse_motion(self.kalman_filter, raw_emb_dists, strack_pool, detections) matches, u_track, u_detection = matching.linear_assignment(dists, thresh=self.match_thresh) ``` Gives much worse results than `fuse_score`:...

Filter out background in bboxes by segmentation before classification

3

Hi @KaiyangZhou. I was wondering if you ever did any experiments regarding the possibility of filtering out the background in the people bounding boxes by using instance segmentation models and...

Store model type and number of classes as metadata in the '.pt' file

3

### 🚀 Feature Request Manual work has to be done to get a custom model working when it could be completely automatized. ```python self.model = models.get( , num_classes=, checkpoint_path =...

Is there a way of defining a string in the hooks args instead of --license-filepath

2

This: ``` - repo: https://github.com/Lucas-C/pre-commit-hooks rev: v1.5.3 hooks: - id: insert-license files: ./* args: - --license-filepath - LICENSE - --use-current-year ``` does not work for me as my LICENSE 600+...

Is there an easy way of evaluating on half of MOT17?

1

I am experimenting quite a lot with trackers and continuously evaluating on MOT17. Even with a multi-gpu setup the evaluation process is very time consuming. Is there an easy way...

How to directly specify sequences to eval and their number of timesteps using `SEQ_INFO`

2

According to: https://github.com/JonathonLuiten/TrackEval/blob/ec237ec3ef654548fdc1fa1e100a45b31a6d4499/trackeval/datasets/mots_challenge.py#L33 It should be possible to specify seq + their associated timesteps such that the `seqinfo.ini` files are not needed. How should the input data look like for...

AttributeError: module 'numpy' has no attribute 'float'

1

In newer numpy versions `np.float` is deprecated. I suggest you change all these: https://github.com/JonathonLuiten/TrackEval/search?q=np.float&type=code For: `np.float64`