Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Releases screenshots for SIMMC-Furniture and API calls with better heuristics (SIMMC v1.2) #60

Closed
wants to merge 15 commits into from
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions .gitattributes
Original file line number Diff line number Diff line change
Expand Up @@ -30,3 +30,7 @@ data/simmc_fashion/fashion_teststd_dials_retrieval_candidates_public.json filter
data/simmc_furniture/furniture_teststd_dials_api_calls.json filter=lfs diff=lfs merge=lfs -text
data/simmc_furniture/furniture_teststd_dials_public.json filter=lfs diff=lfs merge=lfs -text
data/simmc_furniture/furniture_teststd_dials_retrieval_candidates_public.json filter=lfs diff=lfs merge=lfs -text
data/simmc_furniture/furniture_screenshots_part_0.zip filter=lfs diff=lfs merge=lfs -text
data/simmc_furniture/furniture_screenshots_part_1.zip filter=lfs diff=lfs merge=lfs -text
data/simmc_furniture/furniture_screenshots_part_2.zip filter=lfs diff=lfs merge=lfs -text
data/simmc_furniture/furniture_screenshot_map.json filter=lfs diff=lfs merge=lfs -text
2 changes: 2 additions & 0 deletions .lfsconfig
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
[lfs]
fetchexclude = data/simmc_furniture/*.zip
18 changes: 15 additions & 3 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -17,12 +17,19 @@ Our challenge focuses on our SIMMC datasets, both of which are shopping domains:
</figure>



### Latest News

* **[Apr 15, 2021]** Released screenshots for SIMMC-Furniture
([part 0][screenshot_link_0], [part 1][screenshot_link_1], [part 2][screenshot_link_2]).
Also released improved API calls with newer heuristics as SIMMC v1.2 ([PR][screenshot_pr]).
* **[Dec 29, 2020]** Fixed the errors in text spans for both SIMMC-Furniture and SIMMC-Fashion, released new JSON files as SIMMC v1.1 ([PR][span_fix_pr]).
* **[Sept 28, 2020]** Test-Std data released, End of Challenge Phase 1.
* **[June 22, 2020]** Challenge announcement. Training / development datasets are released.
* **[July 8, 2020]** Evaluation scripts and code to train baselines for
Sub-Task #1, Sub-Task #2 released.
* **[July 8, 2020]** Evaluation scripts and code to train baselines for Sub-Task #1, Sub-Task #2 released.
* **[June 22, 2020]** Challenge announcement. Training / development datasets (SIMMC v1.0) are released.

**Note:** DSTC9 SIMMC Challenge was conducted on SIMMC v1.0. Thus all the results and baseline performances are on SIMMC v1.0.


## Important Links

Expand Down Expand Up @@ -149,3 +156,8 @@ SIMMC is released under [CC-BY-NC-SA-4.0](https://creativecommons.org/licenses/b

[dstc9]:https://sites.google.com/dstc.community/dstc9/home
[simmc_arxiv]:https://arxiv.org/abs/2006.01460
[screenshot_link_0]:./data/simmc_furniture/furniture_screenshots_part_0.zip
[screenshot_link_1]:./data/simmc_furniture/furniture_screenshots_part_1.zip
[screenshot_link_2]:./data/simmc_furniture/furniture_screenshots_part_2.zip
[span_fix_pr]:https://github.com/facebookresearch/simmc/pull/54
[screenshot_pr]:https://github.com/facebookresearch/simmc/pull/60
Git LFS file not shown
Git LFS file not shown
4 changes: 2 additions & 2 deletions data/simmc_fashion/fashion_teststd_dials_api_calls.json
Git LFS file not shown
Git LFS file not shown
Git LFS file not shown
3 changes: 3 additions & 0 deletions data/simmc_furniture/furniture_screenshot_map.json
Git LFS file not shown
3 changes: 3 additions & 0 deletions data/simmc_furniture/furniture_screenshots_part_0.zip
Git LFS file not shown
3 changes: 3 additions & 0 deletions data/simmc_furniture/furniture_screenshots_part_1.zip
Git LFS file not shown
3 changes: 3 additions & 0 deletions data/simmc_furniture/furniture_screenshots_part_2.zip
Git LFS file not shown
4 changes: 2 additions & 2 deletions data/simmc_furniture/furniture_teststd_dials_api_calls.json
Git LFS file not shown
4 changes: 4 additions & 0 deletions mm_action_prediction/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,7 @@ We also measure the correctness of the predicted action (API) arguments using at
Specifically, the following API classes and attributes are evaluated.

**SIMMC-Furniture**

| API | API Attributes |
|:------:| :--------: |
| `SearchFurniture` | `furnitureType`, `color` |
Expand All @@ -30,6 +31,7 @@ Each of the above attributes is a categorical variable, modeled as multiclass cl
**Note:** `minPrice` and `maxPrice` attributes corresponding to the `SpecifyInfo` action for Furniture are excluded in the current evaluation.

**SIMMC-Fashion**

| API | API Attributes |
|:--------: | :------: |
| `SearchDatabase` | `attributes` |
Expand Down Expand Up @@ -205,6 +207,8 @@ The baselines trained through the code obtain the following results for Sub-Task
| MN | 81.6 | 1.74 | 61.6 |
| T-HAE | 81.4 | 1.78 | 62.1 |

**Note:** DSTC9 SIMMC Challenge was conducted on SIMMC v1.0. Thus all the results and baseline performances are on SIMMC v1.0.


## Rules for Sub-task #1 Submissions
* Disallowed Input: `belief_state`, `system_transcript`, `system_transcript_annotated`, `state_graph_1`, `state_graph_2`, and anything from future turns.
Expand Down
54 changes: 32 additions & 22 deletions mm_action_prediction/tools/extract_actions.py
Original file line number Diff line number Diff line change
Expand Up @@ -1073,7 +1073,6 @@ def get_roundwise_dialog_actions(subtask, dialog_actions):
# for GET_INFO_ACTION and ADD_TO_CART_ACTION subtask create a
# single reference relative to the carousel
furniture_ids = cur_supervision[ARGS][FURNITURE_ID]
# {'focus': '', 'carousel': ['901712', '1215555']}
if turn_carousel_state["focus"]:
if turn_carousel_state["focus"] in furniture_ids:
cur_supervision[ARGS][FURNITURE_ID] = "focus"
Expand Down Expand Up @@ -1108,15 +1107,20 @@ def get_roundwise_dialog_actions(subtask, dialog_actions):
action_datum["action"] = cur_action
action_datum["action_supervision"] = cur_supervision
action_datum["action_output_state"] = action_output_state
roundwise_actions.append(action_datum)

# Go through all the raw_actions to get the next turn_state.
# Go through all the raw_actions to get the next turn_state
# (only if Share is the last action)
raw_actions = turn_datum["raw_action_with_args"]
if len(raw_actions):
turn_carousel_state = {
"focus": raw_actions[-1][NEXT_STATE][SHARED_FOCUS],
"carousel": raw_actions[-1][NEXT_STATE][SHARED_CAROUSEL],
}
for raw_act in raw_actions:
if raw_act[API] == SHARE:
turn_carousel_state = {
"focus": raw_act[NEXT_STATE][SHARED_FOCUS],
"carousel": raw_act[NEXT_STATE][SHARED_CAROUSEL],
}

# Also record the final carousel_state.
action_datum["final_carousel_state"] = copy.deepcopy(turn_carousel_state)
roundwise_actions.append(action_datum)
return roundwise_actions


Expand All @@ -1131,12 +1135,11 @@ def update_carousel_state(filtered_action, current_state):
insert_carousal_state: Carousel state to insert for the action
new_current_state: State of carousel after executing the action
"""
if filtered_action[API] != GET_INFO_ACTION:
current_state = get_carousel_state(
filtered_action["previousState"], filtered_action
)
# NOTE: For GetInfo, AddToCart, None actions, do not update the states.
ignore_actions = [GET_INFO_ACTION, ADD_TO_CART_ACTION, NONE_ACTION]
# action shouldn't effect the view of the previous state
insert_carousel_state = copy.deepcopy(current_state)
if filtered_action[API] != GET_INFO_ACTION:
if filtered_action[API] not in ignore_actions:
current_state = get_carousel_state(
filtered_action["nextState"], filtered_action
)
Expand Down Expand Up @@ -1174,20 +1177,27 @@ def get_carousel_state(state=None, action_args=None):
return new_state

# Get items in the carousel.
search_order_carousel = ["prefabsInCarousel", "sharedPrefabsInCarousel"]
search_order_focus = ["prefabInFocus", "sharedPrefabInFocus"]
action = action_args["api"]
if action == "SearchFurniture":
new_state["carousel"] = state["sharedPrefabsInCarousel"]
elif action in ["Rotate", "FocusOnFurniture"]:
if action in [
SEARCH_FURNITURE_ACTION,
NAVIGATE_CAROUSEL_ACTION
]:
for order in search_order_carousel:
if state[order]:
new_state["carousel"] = state[order]
# break loop and return - don't return focus item
# SearchFurniture and NavigateCarousel actions
return new_state
elif action in [ROTATE_ACTION, FOCUS_ON_FURNITURE_ACTION]:
focus_id = action_args["args"]["furniture_id"]
search_order = ["prefabsInCarousel", "sharedPrefabsInCarousel"]
for order in search_order:
for order in search_order_carousel:
if focus_id in state[order]:
new_state["carousel"] = state[order]
elif action in ["Previous", "Next"]:
new_state["carousel"] = state["prefabsInCarousel"]

# Focus object.
search_order = ["prefabInFocus", "sharedPrefabInFocus"]
for key in search_order:
for key in search_order_focus:
if state[key] != "":
new_state["focus"] = state[key]
break
Expand Down
2 changes: 2 additions & 0 deletions mm_dst/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -186,6 +186,8 @@ Below is the summary of the [published models](https://github.com/facebookresear
| GPT2 - Fashion (text-only) | 61.2 | 52.1 |
| GPT2 - Fashion (multimodal) | 61.1 | 60.6 |

**Note:** DSTC9 SIMMC Challenge was conducted on SIMMC v1.0. Thus all the results and baseline performances are on SIMMC v1.0.

## Rules for Sub-task #3 Submissions
* Disallowed input per each turn: `belief_state`, `system_transcript`, `system_transcript_annotated`, `state_graph_1`, `state_graph_2`, and anything from future turns.
* If you would like to use any other external resources, please consult with the track organizers (simmc@fb.com). Generally, we allow the use of publicly available pre-trained language models, such as BERT, GPT-2, etc.
Expand Down