bddoia_project's People
Forkers
knightzjz pallekc91 ucsd-svcl arthdh brightyoun scottjingtt khashayarxerx javieraraluce rssharifibddoia_project's Issues
Batch size for "Explainable Object-induced Action Decision for Autonomous Vehicles" 2020 paper
Hello!
In the paper there is no specified batch size in the text that was used for experiments. Could you clarify, was batch_size = 2 used for the experimental results, that can be seen in the paper?
Thank you!
Got Error when train the model.
Hi! @Twizwei ,When I train the model, run
python ./maskrcnn/maskrcnn-benchmark/action_prediction/train.py --batch_size 2 --num_epoch 50 --initLR 0.001 --gtroot "root-to-action-gt" --reasonroot "root-to-explanation-gt" MODEL.SIDE True MODEL.ROI_HEADS.SCORE_THRESH 0.4 MODEL.PREDICTOR_NUM 1 OUTPUT_DIR "output-directory" MODEL.META_ARCHITECTURE "Baseline1"
I got an error " AssertionError: Non-existent key: MODEL.SIDE". I am sure my environment configuration is ok.
Can you give me some advice? Thank you!
Validation and Test Data Missing.
Hi @Twizwei ,
Thanks for sharing this repo :)
I have downloaded lastframe.zip and BDD-OIA.zip as told for data. But I am not finding image data for val and test in there.
As stated in paper, model is trained on 16k images , I found that train image data in lastframe.zip but I can't find 2,270 val images and test set of 4,572 images in those folders. Can you please let me know where could I download them from for evaluation further?
Thanks and regards,
Vaishnavi Khindkar
Training loss isnt decreasing
Im using the faster_rcnn pretrained on BDD100k, and initiated its weight in the model_root parameter. I have set the freeze parameter to True.
I have modified these lines of code in train.py
from line 101,
result = model(imBatch)
pred=result[0]
pred_reason=result[1]
Please help me figure this out. Accumulated training loss remains constant
Freezing FasterRCNN option is set to false by default in code and isn't updated by command line arguments
Hi @Twizwei ,
Freezeing FasterRCNN option is set to false in code and also it's not set to true via arguments. But in paper it's mentioned that faster rcnn is freezed while training. I have changed it's default value to True as of now and have started training again.
Do I need to make any other change to replicate the results?
Thanks,
Vaishnavi Khindkar
The video clip "abf68b89-3b76fa6a.mov" is damaged
it cannot be unzipped. Can you upload this clip to this repo please, the dataset BDD-100k is too large. THX
Training loss goes up and down again
Hi @Twizwei ,
Thanks for sharing the repo.
I have been trying to run this on my local machine but was facing batch size issue so I used gradient accumulation with accumulation_steps as 4 and batch_size = 2. But I see training loss isn't quite decreasing.
I have attached the screenshot for training for 35 epochs. Can you please help with this , like is anything going wrong here?
questions about annotations
Some mistakes about the annotations below:
{"reason": [1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0], "file_name": "dc9226f0-e1cb0a60.jpg"}
The text corresponding to reason one-hot ground truth is Traffic light is green,No lane on the right.
OR Is anything wrong with order of 21 reasons?
Getting same predictions for all the examples in an epoch. Also not predicting explanations at all.
Hi @Twizwei ,
I am getting same predictions for all the examples in an epoch. And getting True only for Forward action even almost after 30 epochs now. And all false predictions for explanations.
Please see the below output for one epoch - I have printed the prediction tensors for actions and explanations. The prediction tensors are changing after many epochs, but for initial many epochs they were almost same for all examples. Can you please help with understanding what's getting wrong here..?
Output sample -
- **********PRed tensor([[ 0.1388, -0.1472, -0.2860, -0.2082],
[ 0.1387, -0.1459, -0.2868, -0.2056]], device='cuda:0')
**********Pred_reason tensor([[-0.6794, -1.6445, -1.2079, -1.2154, -2.5862, -4.5843, -4.7070, -1.1952,
-3.7522, -5.1099, -3.4112, -4.1208, -5.1506, -3.3282, -4.0545, -1.3703,
-1.4235, -1.6865, -1.0422, -1.5427, -2.2461],
[-0.6763, -1.6414, -1.2114, -1.2045, -2.5592, -4.4499, -4.5621, -1.1900,
-3.6692, -4.9239, -3.3501, -4.0153, -4.9572, -3.2591, -3.9511, -1.3657,
-1.4096, -1.6717, -1.0343, -1.5301, -2.2216]], device='cuda:0')
- **********PRed tensor([[ 0.1390, -0.1459, -0.2867, -0.2049]], device='cuda:0')
**********Pred_reason tensor([[-0.6729, -1.6360, -1.2113, -1.2030, -2.5524, -4.4057, -4.5216, -1.1886,
-3.6488, -4.8690, -3.3328, -3.9836, -4.8995, -3.2408, -3.9205, -1.3616,
-1.4094, -1.6679, -1.0320, -1.5294, -2.2124]], device='cuda:0')
prediction action:
tensor([[ True, False, False, False]], device='cuda:0')
ground truth:
[[1. 0. 1. 0.]]
Accumulated Overall Action acc: 0.41973451327433625
prediction reason:
tensor([[False, False, False, False, False, False, False, False, False, False,
False, False, False, False, False, False, False, False, False, False,
False]], device='cuda:0')
ground truth:
[[0. 1. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0.]]
Accumulated Overall Reason acc: 0.0
(2259, 4)
(2259, 4)
Action Random guess acc:[0.52938689 0.47789276 0.35481963 0.37069922]
Action Random guess overall acc:0.3894205646860514
Action Category Acc:[0.70655597 0. 0. 0. ]
Action Average Acc:0.1766389922702548
Action Overall acc:0.41973451327433625
Action f1 macro Acc:[0.35327798 0.35567598 0.42824601 0.41461519]
Action mean f1 macro Acc:0.3879537918791195
Action f1 micro Acc:[0.54625941 0.55201417 0.74900398 0.708278 ]
Action mean f1 micro Acc:0.6388888888888888
Reason Random guess acc:[0.39808917 0.24130879 0.2985258 0.31220096 0.10874704 0.02166667
0.01204819 0.29882207 0.03355705 0.02903501 0.05110733 0.02449694
0.01728608 0.07351713 0.0312229 0.26969124 0.27478937 0.25506073
0.35023041 0.28089129 0.13421829]
Reason Category Acc:[0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
Reason Average Acc:0.0
Reason Overall Acc:0.0
Reason f1 macro Acc:[0.4 0.45540019 0.44153276 0.43369265 0.48377514 0.49755338
0.49855716 0.43595506 0.49587146 0.49766511 0.4925876 0.49688196
0.49822301 0.49087221 0.49710597 0.44359606 0.44632353 0.45670996
0.42431193 0.45183208 0.47744622]
Reason mean f1 macro Acc:0.46742349636736286
Reason f1 micro Acc:[0.66666667 0.83621071 0.79061532 0.76582559 0.93714033 0.99026118
0.99424524 0.77290837 0.98362107 0.99070385 0.97078353 0.98760514
0.99291722 0.96414343 0.98849048 0.79725542 0.8061089 0.84063745
0.73705179 0.82425852 0.91367862]
Reason mean f1 micro Acc:0.8833870865743373
Thank you so much for your time. :)
Regards,
Vaishnavi Khindkar
Can't download your BDD-OIA dataset from your google drive link
Hi all,
Access was denied when I tried to download your dataset using the link below.
https://drive.google.com/open?id=1NzF-UKaakHRNcyghtaWDmc-Vpem7lyQ6
Could you please check this issue?
Thanks,
Hongbo
Cant open Pretrained weights link
Could you please provide the link for pretrained weights. The one showed in the readme is giving 404 error. Thank you
Problem about the action annotation
Hey, in my opinion, the actions could be classified into 5 classes including forward
, stop
, left
, right
and confuse
.
The corresponding one-hot code should be:
[1,0,0,0,0], [0,1,0,0,0], [0,0,1,0,0], [0,0,0,1,0], [0,0,0,0,1]
BUT when I check the validation data, I found a data labeled as [0,0,0,0,0].
What is this?
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.