FEELVOS: Cannot Reproduce the Results

Prerequisites

Please answer the following questions for yourself before submitting an issue.

• I am using the latest TensorFlow Model Garden release and TensorFlow 2.
• I am reporting the issue to the correct repository. (Model Garden official or research directory)
• I checked to make sure that this issue has not already been filed.

1. The entire URL of the file you are using

https://github.com/tensorflow/models/tree/master/research/feelvos

2. Describe the bug

The JF Mean of the model trained by myself is much lower than the pretrained model. Note that both models are trained on DAVIS 2017 ONLY.

3. Steps to reproduce

bash train.sh
bash eval.sh

4. Expected behavior

The JF Mean of the model trained by myself should be the same as the pretrained model.

5. Additional context

The JF Mean of the pretrained model:

#	J-Mean			F-Mean			JF-Mean
0	0.5785237178318143	0.7127632896538271	0.6456435037428208
1	0.9310475299988663	0.9681360145881766	0.9495917722935214
2	0.4347816388873978	0.7517483600500111	0.5932649994687045
3	0.3838198892472484	0.46904564921866604	0.42643276923295725
4	0.8310467622495602	0.8911048883837113	0.8610758253166357
5	0.8821299716976659	0.8492454206504613	0.8656876961740636
6	0.9354415305546544	0.9602273906358517	0.9478344605952531
7	0.937986911770305	0.9676526315903196	0.9528197716803123
8	0.7351716211055481	0.7404584703327636	0.7378150457191559
9	0.9405765388472838	0.9713288672965245	0.9559527030719042
10	0.5058217576563899	0.6123890883698563	0.5591054230131232
11	0.8583363889092226	0.9462034771690471	0.9022699330391348
12	0.9144127878794224	0.8764792475584211	0.8954460177189217
13	0.8651893731262099	0.8811091385792972	0.8731492558527536
14	0.7340012511683769	0.7153217484138518	0.7246614997911143
15	0.7721178723016651	0.9456666171132425	0.8588922447074538
16	0.5674779350946292	0.5620302236001388	0.564754079347384
17	0.7956617403976348	0.8446308466297138	0.8201462935136743
18	0.4179170106668399	0.5344732582195562	0.47619513444319805
19	0.5015225246506723	0.45080785631404624	0.4761651904823593
20	0.7026795627441937	0.8751009883948612	0.7888902755695275
21	0.6704288608954415	0.7362658638598942	0.7033473623776678
22	0.7255354463206185	0.7747192757379563	0.7501273610292873
23	0.7386348935630722	0.7520380420220057	0.7453364677925389
24	0.4711273656173619	0.638404562887137	0.5547659642522494
25	0.9109906698810942	0.9685579814496957	0.939774325665395
26	0.7893043091656985	0.8279587423538309	0.8086315257597647
27	0.6932706223794086	0.8189330099743354	0.756101816176872
28	0.4106005006108749	0.4960857098243246	0.45334310521759974
29	0.6746685667734541	0.7570169959476962	0.7158427813605752
---
All	0.6631594471608241	0.7282175916329557	0.6956885193968899

The JF Mean of the model trained by myself:

#	J-Mean			F-Mean			JF-Mean
0	0.4358827736991943	0.5068412488405826	0.4713620112698884
1	0.9238276759245044	0.975017477431606	0.9494225766780553
2	0.3095491439707915	0.44533813577052395	0.3774436398706577
3	0.24659138880264278	0.2851126127649478	0.2658520007837953
4	0.8388148366242041	0.8636769620155622	0.8512458993198831
5	0.9704355336742194	0.9672196130108436	0.9688275733425316
6	0.904668182964802	0.9621426228975142	0.933405402931158
7	0.9309019439127593	0.9756928188224295	0.9532973813675945
8	0.7811386099174498	0.7649246578725433	0.7730316338949965
9	0.9269253099190823	0.9466916678038179	0.9368084888614501
10	0.28816649407864	0.35886354406196114	0.32351501907030056
11	0.14264716574185424	0.321560958440476	0.23210406209116513
12	0.9276757847105245	0.9247469631290025	0.9262113739197635
13	0.7992973097205738	0.7857453724478146	0.7925213410841943
14	0.5822626783650272	0.5716226237336588	0.576942651049343
15	0.7186079004661688	0.8660500245203817	0.7923289624932752
16	0.506435791453687	0.4830089669869987	0.49472237922034284
17	0.7543824606807855	0.7911582821639339	0.7727703714223597
18	0.3259241472340566	0.4449879951746974	0.38545607120437697
19	0.49586317600886426	0.43534625890136436	0.46560471745511434
20	0.8038883150513086	0.9318701028113064	0.8678792089313075
21	0.6036771675645531	0.69821703697759	0.6509471022710716
22	0.6877390517587146	0.7268843522038472	0.7073117019812809
23	0.5037990523806023	0.5645062597695282	0.5341526560750652
24	0.4135352031535895	0.6028649000251692	0.5082000515893794
25	0.7894040370005461	0.8556991364032608	0.8225515867019034
26	0.7216326650456776	0.7751633829686384	0.7483980240071579
27	0.5721403302867616	0.7346014105728941	0.6533708704298278
28	0.4277384529226039	0.5232603998079526	0.4754994263652782
29	0.4958252152497736	0.6439869751103117	0.5699060951800426
---
All	0.5728746516688414	0.6345614255454813	0.6037180386071614