Plankiformer

Repo Structure

The repo contains the following directories:

• utils: contains auxiliary code.

• out: the output is stored here (no output is uploaded to GitHub, so it must be created).

• Data: the input data is stored here.

• trained-models: contains only best trained models for the end users.

Training models

In order to train a fresh model, use main.py.

The training depends on how the data is structured. There are basically three scenarios:

1. You have directory without the train, test and valid split
2. You have directory already with train and test split
3. The data is in cloud and can be downloaded. Ex: Cifar-10, Stanford-dogs data etc.

For the first case: You can train the model using:

python main.py -datapaths ./data/PhytoData/ -outpath ./out/phyto_out/ -classifier multi -aug -datakind image -ttkind image -save_data yes -resize_images 1 -L 128 -valid_set yes -test_set yes -dataset_name zoolake -training_data False -epochs 40 -finetune 2 -finetune_epochs 40 -balance_weight yes -batch_size 32 -init_name Init_0

There are lots of input commands that can be given to the script. To query them, use the -h flag (python main.py -h).

The performance reports from the ensemble of DeiT-base and EfficientNet-B7 models are shown below for Phytoplankton and Zooplankton images.

1. Ensemble of EfficientNet-B7 models on Phytoplankton images

Accuracy
0.9025121859767529

F1 Score
0.8838444691595628

Classification Report

	precision	recall	f1-score	support
ascomorpha	1	1	1	13
askenasia	0.98	0.98	0.98	59
asplachna	0	0	0	2
asterionella_colonies	1	0.94	0.97	68
asterionella_partial	0.6	0.75	0.67	12
aulacoseira	0.85	0.87	0.86	39
centric_diatoms	0.99	0.99	0.99	158
ceratium	0.97	0.98	0.98	179
chlorophyte	0.85	0.78	0.81	123
chlorophyte_colonial_dividing	0.92	0.85	0.89	172
chlorophyte_elongated	0.85	0.87	0.86	71
chlorophyte_frame	0.93	1	0.96	39
chlorophyte_square	0.9	0.94	0.92	65
chroococcales	0.87	0.96	0.91	27
ciliate_round	0.72	0.84	0.78	25
ciliates	1	0.88	0.94	17
ciliates_blue	0.93	0.96	0.95	85
ciliates_green	0.93	0.95	0.94	60
closterium	0.98	0.98	0.98	128
coelastrum_reticulatum	0.93	0.96	0.95	28
coelosphaerium	0.85	0.9	0.87	68
coleps	1	1	1	10
cosmarium	0.99	0.97	0.98	87
cryptomonas_cryptophyceae	0.94	0.94	0.94	241
cryptophytes_blurry	0.72	0.91	0.81	68
cyanobacteria_colonial_blue	0.9	0.88	0.89	96
cyanobacteria_colonial_clathrate	0.78	0.87	0.82	45
cyanobacteria_colonial_probably	0.94	0.92	0.93	167
cyanobacteria_filamentous	0.95	0.89	0.92	83
didinium	0.69	0.9	0.78	10
dinobryon	0.93	0.96	0.95	170
dinobryon_single_cell	0.88	0.87	0.87	105
dinoflagellate_diamond	0.98	0.96	0.97	57
dolichospermum	0.9	0.99	0.94	93
elakatothrix	0.91	0.98	0.94	59
filament	0.94	0.92	0.93	100
fragilaria	0.98	0.98	0.98	170
gomphosphaeria	1	1	1	2
gonium	0.88	1	0.94	15
gymnodinium	0.93	0.89	0.91	46
hormidium_like	0.94	0.99	0.97	117
kellicottia	1	1	1	6
keratella_cochlearis	0.89	0.89	0.89	55
keratella_quadrata	0.5	0.33	0.4	6
limnoraphis	0.77	0.91	0.83	11
mallomonas_akrokomos	0.62	0.83	0.71	18
mallomonas_big	0.88	0.98	0.93	65
nauplii	0.94	0.94	0.94	18
oocystaceae	0.87	0.95	0.91	80
pandorina	0.93	1	0.96	50
paradileptus	1	0.9	0.95	10
pediastrum	1	1	1	42
pennate_diatom	0.85	0.71	0.77	48
peridinium	0.91	0.92	0.92	92
phacotus	0.8	0.97	0.88	37
plankton_halo	0.93	0.93	0.93	14
plankton_y	1	1	1	7
plankton_z	0.89	1	0.94	24
planktosphaeria	0.8	1	0.89	4
planktothrix	0.91	0.91	0.91	23
polyarthra	0.96	0.88	0.92	56
protist_like_ciliate	0.86	0.94	0.9	129
rhodomonas	0.98	0.9	0.93	88
rotifer	0.52	0.67	0.58	24
rotifer_long	0.75	0.6	0.67	5
rotifer_z	0.93	1	0.96	13
scenedesmus	0.92	0.94	0.93	50
staurastrum	0.98	1	0.99	110
strombidium	0.96	0.93	0.95	87
synchaeta	0.93	0.93	0.93	14
synedra	0.85	0.92	0.88	50
synedra_angustissima	0.92	0.92	0.92	13
synura	0.93	1	0.96	13
tetraedron	1	1	1	26
tintinidium	0.84	0.94	0.89	17
tintinopsis	0.64	1	0.78	7
trichocerca	0.91	0.94	0.93	33
unknown	0.68	0.5	0.58	232
unknown_eccentric	0.67	0.66	0.66	87
unknown_elongated	0.82	0.61	0.7	97
unknown_probably_dirt	0.95	0.92	0.93	114
unrecognizable_dots	0.99	0.98	0.99	152
uroglena	0.93	1	0.96	13
vorticella_epistylis_like	0.86	0.88	0.87	43
zooplankton	0.82	0.92	0.87	72
accuracy			0.9	5334
macro avg	0.87	0.9	0.88	5334
weighted avg	0.9	0.9	0.9	5334

2. Ensemble of EfficientNet-B7 models on Zooplankton images

Accuracy
0.9376271186440678

F1 Score
0.8956439313243028

Classification Report

	precision	recall	f1-score	support
aphanizomenon	0.94	0.96	0.95	48
asplanchna	0.97	1	0.99	102
asterionella	0.99	1	0.99	158
bosmina	1	0.92	0.96	13
ceratium	0.99	0.98	0.98	155
chaoborus	1	1	1	2
collotheca	0.93	1	0.96	38
conochilus	1	1	1	40
copepod_skins	0.71	1	0.83	5
cyclops	0.93	0.96	0.95	300
daphnia	0.95	0.98	0.97	296
daphnia_skins	0.67	0.84	0.74	19
diaphanosoma	1	0.98	0.99	175
diatom_chain	0.67	1	0.8	2
dinobryon	0.96	0.99	0.97	550
dirt	0.55	0.3	0.39	20
eudiaptomus	0.96	0.92	0.94	231
filament	0.95	1	0.98	61
fish	0.8	0.98	0.88	46
fragilaria	0.98	0.98	0.98	196
hydra	1	1	1	3
kellicottia	0.97	1	0.99	78
keratella_cochlearis	0.86	0.84	0.85	117
keratella_quadrata	0.95	0.98	0.96	131
leptodora	0.93	0.93	0.93	41
maybe_cyano	0.96	1	0.98	205
nauplius	0.97	0.95	0.96	391
paradileptus	0.99	0.99	0.99	87
polyarthra	0.73	0.58	0.65	19
rotifers	0.8	0.84	0.82	166
synchaeta	0.93	0.95	0.94	56
trichocerca	0.9	0.99	0.94	86
unknown	0.77	0.65	0.71	249
unknown_plankton	0.52	0.3	0.38	46
uroglena	1	1	1	293
accuracy			0.94	4425
macro avg	0.89	0.91	0.9	4425
weighted avg	0.93	0.94	0.93	4425

3. Ensemble of DeiT-B models on Zooplankton images

Accuracy
0.9340112994350283

F1 Score
0.8781533144325483

Classification Report

	precision	recall	f1-score	support
aphanizomenon	0.88	0.94	0.91	48
asplanchna	0.96	1	0.98	102
asterionella	0.99	0.99	0.99	158
bosmina	1	0.92	0.96	13
ceratium	0.99	0.96	0.97	155
chaoborus	1	1	1	2
collotheca	0.9	1	0.95	38
conochilus	1	1	1	40
copepod_skins	0.75	0.6	0.67	5
cyclops	0.92	0.96	0.94	300
daphnia	0.93	0.98	0.95	296
daphnia_skins	0.7	0.74	0.72	19
diaphanosoma	1	0.98	0.99	175
diatom_chain	1	0.5	0.67	2
dinobryon	0.95	0.99	0.97	550
dirt	0.57	0.2	0.3	20
eudiaptomus	0.96	0.92	0.94	231
filament	0.91	1	0.95	61
fish	0.88	0.93	0.91	46
fragilaria	0.99	0.98	0.99	196
hydra	1	1	1	3
kellicottia	0.97	1	0.99	78
keratella_cochlearis	0.87	0.85	0.86	117
keratella_quadrata	0.95	0.98	0.96	131
leptodora	0.88	0.9	0.89	41
maybe_cyano	0.97	1	0.98	205
nauplius	0.96	0.97	0.96	391
paradileptus	0.99	0.99	0.99	87
polyarthra	0.65	0.58	0.61	19
rotifers	0.81	0.86	0.83	166
synchaeta	0.96	0.96	0.96	56
trichocerca	0.92	0.97	0.94	86
unknown	0.74	0.62	0.68	249
unknown_plankton	0.48	0.24	0.32	46
uroglena	1	1	1	293
accuracy			0.93	4425
macro avg	0.9	0.87	0.88	4425
weighted avg	0.93	0.93	0.93	4425