Skip to content

Commit

Permalink
no crash resnet 34
Browse files Browse the repository at this point in the history
  • Loading branch information
felipecode committed Apr 16, 2019
1 parent 3bfafaa commit 31924cd
Show file tree
Hide file tree
Showing 11 changed files with 26 additions and 17 deletions.
1 change: 1 addition & 0 deletions configs/nocrash/resnet34imnet10-nospeed.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@
SAVE_SCHEDULE: range(0,1000001, 2000) # The iterations where training checkpoints are going to be saved
NUMBER_OF_LOADING_WORKERS: 12 # Number of threads used in the data loader
MAGICAL_SEED: 26957017
FINISH_ON_VALIDATION_STALE: 'TrainValidation'
#### Input related parameters ####
# A dictionary with all the sensors that are going to be used as input
# this should match the train dataset
Expand Down
1 change: 1 addition & 0 deletions configs/nocrash/resnet34imnet100-nospeed.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@
SAVE_SCHEDULE: range(0,1000001, 2000) # The iterations where training checkpoints are going to be saved
NUMBER_OF_LOADING_WORKERS: 12 # Number of threads used in the data loader
MAGICAL_SEED: 26957017
FINISH_ON_VALIDATION_STALE: 'TrainValidation'
#### Input related parameters ####
# A dictionary with all the sensors that are going to be used as input
# this should match the train dataset
Expand Down
1 change: 1 addition & 0 deletions configs/nocrash/resnet34imnet100.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@
SAVE_SCHEDULE: range(0,1000001, 2000) # The iterations where training checkpoints are going to be saved
NUMBER_OF_LOADING_WORKERS: 12 # Number of threads used in the data loader
MAGICAL_SEED: 26957017
FINISH_ON_VALIDATION_STALE: 'TrainValidation'
#### Input related parameters ####
# A dictionary with all the sensors that are going to be used as input
# this should match the train dataset
Expand Down
1 change: 1 addition & 0 deletions configs/nocrash/resnet34imnet10S1.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@
SAVE_SCHEDULE: range(0,1000001, 2000) # The iterations where training checkpoints are going to be saved
NUMBER_OF_LOADING_WORKERS: 12 # Number of threads used in the data loader
MAGICAL_SEED: 26957017
FINISH_ON_VALIDATION_STALE: 'TrainValidation'
#### Input related parameters ####
# A dictionary with all the sensors that are going to be used as input
# this should match the train dataset
Expand Down
1 change: 1 addition & 0 deletions configs/nocrash/resnet34imnet10S2.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@
SAVE_SCHEDULE: range(0,1000001, 2000) # The iterations where training checkpoints are going to be saved
NUMBER_OF_LOADING_WORKERS: 12 # Number of threads used in the data loader
MAGICAL_SEED: 26957017
FINISH_ON_VALIDATION_STALE: 'TrainValidation'
#### Input related parameters ####
# A dictionary with all the sensors that are going to be used as input
# this should match the train dataset
Expand Down
19 changes: 12 additions & 7 deletions docs/exploring_limitations.md
Original file line number Diff line number Diff line change
@@ -1,7 +1,6 @@
Exploring the Limitations of Behavior Cloning for Autonomous Driving
====================================================================

### Reproducing the results

#### Downloading Checkpoints

Expand All @@ -14,18 +13,24 @@ all the needed files.
python3 tools/download_nocrash_models.py


#### Models present
#### Models available

* *resnet34imnet10S1*: is the model with the random seed 1 from Figure 6 and it is also our best model
* *resnet34imnet10S2*: is the model with the random seed 2 from Figure 6
* *resnetmodel without the speed prediction and ten hours of training (Yellow model Fig. 5)
*
* *resnet34imnet10S1*: is the model with the random seed 1 from Figure 6
* *resnet34imnet10S2*: is the model with the random seed 2 from Figure 6 and it is also our best model (Green Model Fig. 5)
* *resnet34imnet10-nospeed*: without the speed prediction and ten hours of training (Yellow model Fig. 5)
* *resnet34imnet100*: the model with 100 hours of demonstrations (Blue model Fig. 5)
* *resnet34imnet100-nospeed*: the model with 100 hours of demonstrations and no speed prediction (Red model Fig. 5)



#### Reproducing the results

#### View our best model driving
To reproduce one any of these models run:

python3 coiltraine.py -de


To test all of them in batch:



Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -16,10 +16,10 @@



class CARLA100NewTown(ExperimentSuite):
class NocrashNewTown(ExperimentSuite):

def __init__(self):
super(CARLA100NewTown, self).__init__('Town02')
super(NocrashNewTown, self).__init__('Town02')

@property
def train_weathers(self):
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -13,10 +13,10 @@
from carla.settings import CarlaSettings
from carla.driving_benchmark.experiment_suites.experiment_suite import ExperimentSuite

class CARLA100NewWeather(ExperimentSuite):
class NocrashNewWeather(ExperimentSuite):

def __init__(self):
super(CARLA100NewWeather, self).__init__('Town01')
super(NocrashNewWeather, self).__init__('Town01')

@property
def train_weathers(self):
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -16,10 +16,10 @@



class CARLA100NewWeatherTown(ExperimentSuite):
class NocrashNewWeatherTown(ExperimentSuite):

def __init__(self):
super(CARLA100NewWeatherTown, self).__init__('Town02')
super(NocrashNewWeatherTown, self).__init__('Town02')

@property
def train_weathers(self):
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -14,12 +14,11 @@
from carla.driving_benchmark.experiment_suites.experiment_suite import ExperimentSuite


# TODO: maybe add aditional tasks ( NO dynamic obstacles for instace !)

class CARLA100Training(ExperimentSuite):
class NocrashTraining(ExperimentSuite):

def __init__(self):
super(CARLA100Training, self).__init__('Town01')
super(NocrashTraining, self).__init__('Town01')

@property
def train_weathers(self):
Expand Down
2 changes: 1 addition & 1 deletion tools/download_nocrash_models.py
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@

# Download the full data from the models
print ("Downloading the coil models checkpoints 1.1 GB")
file_id = '1ISOaTjMS-c-73krlIewCLuneTjNxz5XI'
file_id = '1U4GpTMUN2PWpTQd96gDP2XGVk4OwzONg'
destination_pack = 'nocrash_basic.tar.xz'

download_file_from_google_drive(file_id, destination_pack)
Expand Down

0 comments on commit 31924cd

Please sign in to comment.