3 classes

[hassane11]

HI there, I was testing your model for the 3 classes using the images you provided...I tested it several times and the best accuracy was 80%... I even got better results using Densenet..

I took random images and from the dataset and set them as validation... am i doing something wrong?

[muhammedtalo]

Hi, I have provided the notebook for also three classes. Please see COVID-19 main repository.

[JakiaSultana]

HI there, I was testing your model for the 3 classes using the images you provided...I tested it several times and the best accuracy was 80%... I even got better results using Densenet..

I took random images and from the dataset and set them as validation... am i doing something wrong?

Hello, Hassane11..I'm Jenny from Bangladesh. I'm trying to run this model but somehow I'm facing some difficulties..would you please help me to work it out? please give me your email address to reach you.

[yaelbeng]

Hi, I run your notebook for three classes many times and got COVID-19 recall in the range of 0.7-0.78 the most. Is it possible that something is missing in the notebook? I really appreciate your help.

[muhammedtalo]

Please install the "fastai" (v1) and use the "one-fit-cycle"

let us know if you get the desired results.

[yaelbeng]

I run the notebook on the colab, I did not make any changes to the code. I used fastai version =1.0.61 , torch version =1.4 and torchvision version =0.5.0. I attached the colab notebook after I downloaded the data to a folder called dataset. https://drive.google.com/file/d/1U9_w1U59yQAn9xEqTZ7ZrifMXY2NJCMK/view?usp=sharing Is there anything I need to change? Thanks.

These are the results:

225 188 0.8355555555555556 [[21 3 3] [ 0 91 17] [ 0 14 76]] precision recall f1-score support

Covid-19 1.00 0.78 0.88 27 No_findings 0.84 0.84 0.84 108 Pneumonia 0.79 0.84 0.82 90

accuracy                           0.84       225

macro avg 0.88 0.82 0.84 225 weighted avg 0.84 0.84 0.84 225

epoch	train_loss	valid_loss	accuracy	time
0	0.802216	1.120054	0.146667	03:18
1	0.719557	0.613483	0.733333	00:49
2	0.657062	0.565587	0.746667	00:50
3	0.616445	0.561667	0.720000	00:49
4	0.597185	0.604905	0.755556	00:48
5	0.595375	0.593846	0.751111	00:48
6	0.551170	0.499635	0.800000	00:48
7	0.528017	0.564676	0.768889	00:49
8	0.522138	0.666555	0.702222	00:48
9	0.498333	0.450894	0.808889	00:48
10	0.506787	0.859526	0.684444	00:48
11	0.506639	0.832571	0.724444	00:49
12	0.510909	1.550252	0.524444	00:48
13	0.509355	0.843340	0.622222	00:47
14	0.510913	0.761859	0.755556	00:47
15	0.524859	1.158196	0.604444	00:48
16	0.527368	0.469246	0.831111	00:49
17	0.550249	1.834808	0.471111	00:48
18	0.548067	0.792381	0.666667	00:48
19	0.546760	0.573853	0.742222	00:48
20	0.546302	0.621358	0.724444	00:48
21	0.555524	0.745503	0.737778	00:48
22	0.540617	1.023847	0.653333	00:49
23	0.536341	0.667747	0.715556	00:49
24	0.549241	0.595138	0.755556	00:49
25	0.525991	0.646543	0.755556	00:49
26	0.528114	2.888739	0.431111	00:49
27	0.522915	1.264647	0.586667	00:50
28	0.523076	0.820289	0.671111	00:49
29	0.519192	0.934004	0.617778	00:49
30	0.511381	0.672165	0.706667	00:50
31	0.518954	1.238636	0.551111	00:50
32	0.534379	0.647150	0.737778	00:50
33	0.498529	0.499963	0.826667	00:49
34	0.511284	0.718220	0.715556	00:49
35	0.519870	0.530848	0.808889	00:49
36	0.496464	1.174538	0.640000	00:49
37	0.492945	0.631620	0.746667	00:49
38	0.478762	0.654109	0.720000	00:50
39	0.486948	0.843268	0.648889	00:50
40	0.478552	0.560477	0.777778	00:49
41	0.479822	0.517507	0.791111	00:49
42	0.477410	1.204803	0.462222	00:48
43	0.472072	0.764760	0.697778	00:48
44	0.457440	0.539294	0.791111	00:48
45	0.454724	0.534468	0.800000	00:49
46	0.464955	1.174639	0.555556	00:49
47	0.468718	1.059415	0.497778	00:49
48	0.454491	1.226268	0.640000	00:48
49	0.440996	0.494795	0.826667	00:49
50	0.441115	0.492688	0.813333	00:50
51	0.434957	0.565116	0.786667	00:50
52	0.416702	0.671017	0.777778	00:50
53	0.410208	0.797799	0.733333	00:50
54	0.397721	0.500998	0.826667	00:50
55	0.403954	0.932937	0.617778	00:49
56	0.402539	0.552333	0.791111	00:49
57	0.392269	0.856246	0.671111	00:49
58	0.393315	0.508536	0.804444	00:50
59	0.398827	0.539836	0.800000	00:49
60	0.392190	0.617230	0.777778	00:50
61	0.383788	0.462455	0.791111	00:50
62	0.375399	0.667062	0.706667	00:50
63	0.363777	0.595107	0.768889	00:50
64	0.363340	0.530078	0.773333	00:49
65	0.363703	0.510627	0.786667	00:50
66	0.364770	0.585928	0.777778	00:49
67	0.365704	0.485989	0.800000	00:50
68	0.346458	0.455922	0.826667	00:50
69	0.340564	0.687559	0.715556	00:51
70	0.344715	0.583831	0.777778	00:51
71	0.336572	0.678242	0.782222	00:50
72	0.324699	0.628685	0.782222	00:50
73	0.316564	0.438614	0.826667	00:50
74	0.311029	0.514318	0.795556	00:51
75	0.309957	0.543917	0.804444	00:50
76	0.293605	0.579203	0.813333	00:50
77	0.293663	0.446774	0.844444	00:50
78	0.295200	0.499016	0.817778	00:50
79	0.283304	0.535741	0.813333	00:50
80	0.279290	0.472687	0.835556	00:50
81	0.277194	0.455442	0.848889	00:50
82	0.277741	0.472730	0.813333	00:50
83	0.275161	0.452817	0.848889	00:50
84	0.281002	0.478777	0.808889	00:50
85	0.276415	0.447774	0.835556	00:50
86	0.272619	0.447729	0.848889	00:50
87	0.268938	0.441842	0.853333	00:49
88	0.268014	0.478940	0.835556	00:50
89	0.280847	0.480289	0.831111	00:50
90	0.274180	0.438755	0.853333	00:50
91	0.272975	0.446968	0.848889	00:50
92	0.265561	0.445463	0.853333	00:50
93	0.257683	0.445405	0.848889	00:50
94	0.256960	0.453300	0.844444	00:50
95	0.253767	0.450355	0.844444	00:50
96	0.245825	0.444155	0.848889	00:50
97	0.251146	0.445887	0.844444	00:51
98	0.244173	0.446138	0.848889	00:50
99	0.246019	0.449752	0.835556	00:50

[muhammedtalo]

Way to go! You have improved your previous score. In our paper we have reported five fold cross validation evaluation. For Fold-3, we have reported a classification accuracy of 83.6% . Your result is better than that.

For the next task, you can do 5-fold cross validation and see how the results in other folds or you can change the seed number (np.random.seed(41)) in the code and see how the results change for different distributions.

Would you also share your binary results. Best

[yaelbeng]

Hi, thanks for your answer. My problem was with the recall, for all 5-fold cross validation the COVID-19 recalls that were reported in the paper are higher than 0.94, and my result was significantly lower. Changing the seed number helps. Thank you.

[minaahmed]

Hello, How did you split the dataset for three classes code?

[jaideep11061982]

@minaahmed do you know if all the models https://github.com/lindawangg/COVID-Net/blob/master/docs/models.md

are 3 classs models only?

One more q. Can this model detect all kind of opacities or Pneumonia opacity and Covid Opacity only ?

[mohanades]

Hi muhammed I have a problem in first run.When I run your source I have a problem I attach that message. How can I upload dataset? Can I have first of your source code? Thank you