I use a similar setup with the mapillary_tools scripts.

For setting the compass angle only, specify --overwrite_EXIF_direction_tag in the image processing step below but don't specify --overwrite_EXIF_gps_tag.

My work flow for geo-tagging photos and uploading them to mapillary is processing and uploading is usually the following (on Linux):

Current version of mapillary_tools

Installation

Run the following commands for installing mapillary_tools:

• sudo apt install python-pip
• pip install git+https://github.com/mapillary/Piexif
• pip install --upgrade git+https://github.com/mapillary/mapillary_tools

Image Processing

Run the following command for adding GPS information to your photo's EXIF tags:

~/.local/bin/mapillary_tools process --verbose --advanced --rerun --user_name mapillary_user --interpolate_directions --overwrite_EXIF_gps_tag --overwrite_EXIF_direction_tag --duplicate_angle 360 --offset_time 1 --import_path "~/media/cam/" --geotag_source "gpx" --geotag_source_path "~/gpx/today.gpx"

Some explanations:

• Replace mapillary_user with your mapillary user name
• Replace ~/media/cam/ and ~/gpx/today.gpx with your photo and GPX path
• --interpolate_directions tells it to interpolate the photo direction based on the GPX track
• --duplicate_angle 360 tells it to ignore the angle when searching for duplicates, usefule since without GPS support your camera won't set a meaningful angle
• --offset_time 1 specifies the offset to add to your GPX time stamps to synchronize with the photo timestamp from your camera clock

You can run mapillary_tools process --advanced -h to see all options and their description.

Upload

Run the following command to upload your photos:

~/.local/bin/mapillary_tools upload --verbose --advanced --skip_subfolders --number_threads 10 --max_attempts 10 --import_path "~/media/cam/"

Old answer for old version of mapillary_tools (before release v0.0.0)

export MAPILLARY_USERNAME=myname
export MAPILLARY_EMAIL=myaddress@domain.com
export MAPILLARY_PASSWORD=asdf1234

python geotag_from_gpx.py --time-offset 0 /path/to/images/ /path/to/track.gpx
python add_mapillary_tag_from_exif.py /path/to/images/
python remove_duplicates.py -p 360 /path/to/images/
python upload.py /path/to/images/

• geotag_from_gpx.py adds the coordinates for each photo and also the heading based on the next photo. This is the most interesting script. You can use the --time-offset parameter if your camera's clock is out of sync with your GPS logger's clock. Ideally use JOSM afterwards to open a few photos and validate the geotagging.
• add_mapillary_tag_from_exif.py adds a special tag to identify the photo sequence.
• remove_duplicates.py removes duplicated photos taken at the same location. (Parameter -p 360 is important for ignoring the heading when searching for duplicates. This is because the script tries to keep photos at the same location if the heading changes significantly. But due to the previously executed geotag_from_gpx.py the heading for photos taken at the same location is almost random, leading to large heading variations and thus many duplicates being erroneously kept).
• upload.py will simply upload the result to mapillary.

I set it with the python script interpolate_direction.py from Mapillary. You can also use Javawa's fotogeotag

After uploading your sequence, once you get the notification it's available on Mapillary, edit sequence and click on NORMALIZE (it will adjust the bearing of the photo in relation to the next one). No script needed.

Also for geotag I recommend GeoSetter, you can adjust timestamp with positive and negative values (which is extremely useful as you cannot sync HH:MM:SS between yours GPS/smartphone and the action camera (unless it's adjustable by the action cam app, as in GoPro Hero 5 or GoPro Session 5).