Neural networks trained on datasets such as ImageNet have led to major advances in visual object classification. One obstacle that prevents networks from reasoning more deeply about complex scenes and situations, and from integrating visual knowledge with natural language, like humans do, is their lack of common sense knowledge about the physical world. Videos, unlike still images, contain a wealth of detailed information about the physical world. However, most labelled video datasets represent high-level concepts rather than detailed physical aspects about actions and scenes. In this work, we describe our ongoing collection of the "something-something" database of video prediction tasks whose solutions require a common sense understanding of the depicted situation. The database currently contains more than 100,000 videos across 174 classes, which are defined as caption-templates. We also describe the challenges in crowd-sourcing this data at scale.

No abstract

HFRCB in combination with weekly cisplatin achieves a high rate of locoregional control and survival. Four weekly cycles of 40 mg/m2 cisplatin seem to be the dose limit for most patients.

The increased knowledge regarding HPV-infections in head and neck squamous cell carcinoma (HNSCC) has unexpectedly contributed to several uncertainties related to i) prevalence diversities depending on tumour site and geographical origin of the patients, ii) proportion of HPV-driven tumours among HPV-DNA-positive cases, and iii) identification of patients with HPV-attributed survival benefit. To investigate this heterogeneity, we analysed 307 HNSCC cases (tonsillar, n=135; non-tonsillar, n=172) from eight health care centers mostly from Northern Germany and determined HPV-DNA/mRNA and p16INK4A-status and combined results with the patient outcome. Overall HPV-DNA prevalence rate was 23.5% (72/307); attributed to: 43.7% (59/135) and 7.6% (13/172) tonsillar and non-tonsillar cases, respectively. Among these, 96.6% tonsillar and 38.5% non-tonsillar SCC were HPV-mRNA-positive. Although the study cohort was composed of patients from regions of rather close proximity, prevalence rates showed diversities of up to 40% in HNSCC subsite analysis with the lowest prevalence for tonsillar SCC in metropolitan areas (22.2%) vs. 50.9% in rural areas. Survival analysis identified p16INK4A alone as strongest predictor, followed by HPV-DNA-status alone or in combination with p16INK4A. This survival benefit was shown for tonsillar and non-tonsillar cases. Smoking significantly correlated with HPV-status, however, it does not influence survival when stratified for HPV. In conclusion, the data emphasize the urge for further data on HPV-infection in HNSCC to, e.g. clarify to what extent survival benefits of p16INK4A-positive patients are truly attributed to HPV-infections.

A prospective multicenter phase-II trial (12 centers) was performed by the German larynx organ preservation group (DeLOS) to evaluate the effect of induction chemotherapy (ICHT) with paclitaxel/cisplatin (TP), followed by accelerated-hyperfractionated (concomitant boost) radiotherapy (RT) in responders. The trial was focused on larynx preservation, tumor control, survival, salvage surgery and late toxicity in patients with advanced larynx/hypopharynx carcinoma eligible for total laryngectomy (LE). Seventy-one patients (40 larynx, 87.5% St. III, IV; 31 hypopharynx, 93.4% St. III, IV) were enrolled into the study and treated with ICHT (200 mg/m(2) paclitaxel, 100 mg/m(2) cisplatin; day 1, 22) according to the DeLOS protocol. Patients with complete or partial tumor response proceeded to RT (69.9 Gy in 5.5 weeks). Non-responders received a LE followed by postoperative RT (56-70 Gy in 5.5-7 weeks). The response rate to ICHT for larynx cancer was 69.6% (7.1% complete, 62.5% partial response) and for hypopharyngeal cancer was 84.3% (6.9% complete, 77.4% partial response). Overall survival after 36 months was 60.3% (95% CI, 48.4-72.2%), after 42 months was 56.5% (95% CI, 44.2-68.8%). Laryngectomy-free survival was as follows: after 36 months, 43.0% (95% CI, 30.9-55.0%); after 42 months, 41.3% (95% CI, 29.3-53.3%). Both parameters did not show different outcomes after distinguishing larynx from hypopharynx. LE was indicated in 15 non-responders after ICHT. Five of the 15 non-responders refused the laryngectomy. Two of the five received RT instead and had no evidence of disease 42 months after RT. Late toxicity (dysphagia III, IV LENT SOMA score in laryngectomy-free survivors: after 6 months, 1.8%; 12 months, 11.4%; 18 months, 14.5%; 24 months, 8.1%; 36 months, 16%) and salvage surgery (4 pharyngocutaneous fistulas in 27 operations) were tolerable. In a large portion of patients eligible for LE, the larynx could be preserved with satisfying functional outcome. Good responders after ICHT had also a good general outcome with relatively rare severe late toxicities. Due to a slight increase of relevant late dysphagia, functional outcome regarding swallowing and tracheotomy free breathing should be more focused in future larynx organ preservation trials.