Table 1 Summary of the top solutions

From: Analysis of the Human Protein Atlas Weakly Supervised Single-Cell Classification competition

Rank

Team name

mAP score

No. of submissions

Overview of methods

1

bestfitting

0.5667

480

Fair cell activation network, rule-based cell labeling based on image-level predictions, transformer

2

[red.ai]

0.55328

459

Cell-level models, image-level models, custom loss

3

MPWARE & ZFTurbo & Dieter

0.54995

500

Puzzle-CAM, image-level models, out-of-fold predictions

4

MILIMED

0.54389

258

Data-centric approach, manual labels, cell-level models

5

narsil & David & tito

0.54243

518

Image-level model predicts images made up of augmentations of one cell

6

scumed&Mitotic Spindle

0.54108

435

Image-level models, cell-level models

7

PFCell

0.54053

162

Image-level models, CAM for cells, pseudo-labels

8

Guanshuo Xu

0.53898

62

Image-level label multiplied with CAMs

9

AllDataAreExt & Galixir

0.53557

488

Cell-level models, image-level models

10

[RAPIDS.AI] Cell Game [Rist]

0.53503

166

Cell-level models, manual label for rare class, treat weak label as noise

11

Silvers

0.53287

411

Cell-level models, image-level models, pseudo-labels

12

Andrew Tratz

0.53049

134

Image-level models, cell-level models

13

CVSSP + forecom.ai

0.52717

202

Manual label for rare class, image-level models, cell-level models

15

Fumihiro Kaneko

0.5186

144

Image-level models, pesudo-labels from CAM

16

Looking for the lost cell

0.51711

431

Pseudo-labels, image-level models, cell-level models

18

yuvaramsingh

0.51559

162

Image-level labels, transformer

20

Da Yu

0.50626

88

Cell-level models, image-level labels

21

Alexander Riedel

0.50249

238

Image-level models multiplied by CAMs

22

cool-rabbit

0.50106

399

Pseudo-labels, cell-level models

23

Raman

0.50016

150

Pseudo-labels, gradient accumulation

24

Shai

0.49222

209

Image-level model multiplied by CAMs

27

Mikhail Gurevich

0.48319

85

Cell-level models (Puzzle-CAM), image-level models, pseudo-labels

28

Histopathological Challenger

0.48066

26

Cell-level models, image-level models, pseudo-labels

32

Quoc-Hung To

0.46893

152

Cell-level models, image-level models